Download Labor Mobility and the Cross-Section of Expected Returns

Document related concepts

Modern portfolio theory wikipedia , lookup

Transcript
Labor Mobility and the Cross-Section of Expected
Returns∗
Andrés Donangelo
Haas School of Business
University of California, Berkeley†
This version: January 3, 2011
Job Market Paper
Abstract
Worker’s employment decisions affect the productivity of capital and asset prices
in predictable ways. Using a dynamic model, I show that reliance on a workforce
with flexibility to enter and exit an industry translates into a form of operating
leverage that amplifies equity-holders’ exposure to productivity shocks. Consequently,
firms in an industry with mobile workers have higher systematic risk loadings and
higher expected asset returns. I use data from the Bureau of Labor Statistics to
construct a novel measure of labor supply mobility, in line with the model, based on
the composition of occupations across industries over time. I document a positive and
economically significant cross-sectional relation between labor mobility and expected
asset returns. This relation is not explained by firm characteristics known in the
literature to predict expected returns.
∗
I would like to thank Nicolae Gârleanu, Robert Anderson, Maria Cecilia Bustamante, Esther Eiling,
Hayne Leland, Martin Lettau, Dmitry Livdan, Ulrike Malmendier, Miguel Palacios, Christine Parlour,
Mark Rubinstein, Adam Szeidl, Johan Walden, and participants at Haas School of Business lunch seminars,
U.C. Berkeley Financial Economics seminar, and LBS Trans-Atlantic Doctoral Conference for helpful
comments. Errors are solely my responsibility. I gratefully acknowledge the Dean Witter Foundation, the
White Foundation, and the NASDAQ OMX Educational Foundation for financial support.
†
Contact information: 545 Student Services Bldg., Berkeley CA 94720-1900; e-mail: [email protected], http://faculty.haas.berkeley.edu/donangel/donangelo.html
1
1
Introduction
Factor mobility–the ability to reallocate resources across the economy–maximizes aggregate
output, mitigates asymmetric shocks, and thus increases total welfare. Although a positivesum game at the aggregate level, factor mobility makes some players worse off, in particular
those who lack control over it. I focus here on a type of factor mobility risky for the owners
of a firm and relatively unexplored by the literature until now: labor mobility.
This paper addresses the following question: are firms that employ workers with flexibility to move across industries more exposed to systematic risk than firms in which workers
have low flexibility to switch industries? The answer to this question is important for a
number of reasons. First, it helps us to understand where a firm’s loading on systematic
risk comes from by linking workers’ decisions to asset-price dynamics. Second, it sheds
light on why observable properties of the workforce employed in an industry should predict expected asset returns. Finally, it motivates empirical evidence that suggests that
labor mobility represents an economically significant mechanism for asset pricing: zero-net
investment portfolios that hold long positions in stocks of high-mobility firms and short
positions in stocks of low-mobility firms earn an annual return spread of up to 5.5%, after
controlling for firm characteristics known to predict expected stock returns.
I establish the link between workers’ employment decisions and expected asset returns
in a dynamic model of an industry where firms face a mobile labor supply. Labor mobility
is determined by the nature of labor skills required by a productive technology common
to all firms in the industry. The more specific the labor skills required by occupations
in the industry, the less mobile labor supply is. I first present a simple version of the
model that conveys the main intuition of the labor-mobility mechanism. The simple model
considers the special extreme cases of immobile and fully mobile labor. The level of labor
employed in the industry is static in the immobile labor case, while it depends on the
relative performance of the industry with respect to the economy in the fully mobile case.
The model predicts that when economy-wide wages are sufficiently smooth relative to the
industry’s productivity shocks, reliance on a mobile labor supply translates into a form of
operating leverage and increases a firm’s exposure to systematic risk.
The basic idea of the labor mobility mechanism can be summarized as follows: Firms are
continuously competing for workers in labor markets. Some industries, such as Health Care,
are associated with segmented labor markets where workers have high levels of industryspecific labor specialization. Competition for workers in these “immobile” industries is
2
limited to firms within the industry. Other industries, such as Retail Trade, rely on broader
labor markets, where workers have more general skills. Workforces in these industries are
mobile, since they are able to search for higher wages across broader sectors, and thus
are less affected by industry-specific shocks. Firms in “mobile” industries compete for
workers with firms in other industries as well, which leads to labor costs per worker that
are less affected by the industry-specific performance and thus lever up the firm’s exposure
to systematic risk.
Asset-pricing implications of labor mobility depend on the level of cyclicality and labor
intensity of the industry. A firm in a mobile industry is more able to attract workers when
its performance is relatively high with respect to that of the economy. Industries with
output highly correlated to the business cycle tend to attract workers in good times and
lose workers in bad times. The pro-cyclicality of labor supply amplifies the pro-cyclicality of
productivity of capital, and attenuates the pro-cyclicality of wages in the industry. These
effects are increasing in the labor intensity–the weight of labor on total output–of the
industry.
I also consider intermediate levels of labor mobility. The richer setting is based on
the interaction between two distinct labor markets: an industry-specific labor market of
specialists in the industry and an economy-wide labor market of generalists. I model labor
mobility as the ability of specialists and generalists to move across these two labor markets. I provide closed-form solutions for the value of the firm and instantaneous expected
asset returns. I show that under the plausible assumptions, expected asset returns are
monotonically increasing in labor mobility.
Industry-specific labor mobility is not directly observable and represents a challenge for
the study of factor mobility. To overcome this problem, I propose a new indirect measure
of labor mobility based on the occupational composition across industries. The measure
represents the level of across-industry concentration of occupations, based on data from
the Bureau of Labor Statistics from 1988 to 2009. Workers in occupations concentrated
in fewer industries are associated with industry-specialists with low labor mobility, while
workers in occupations dispersed across the economy are associated with generalists with
high mobility. The measure of inter-industry labor mobility captures the mean occupation
dispersion of employed workers in a given industry, weighted by the number of workers
assigned to each occupation.
I present empirical support for a positive relation between labor mobility and expected
asset returns. Firm- and industry-level portfolios of stocks sorted on the measure of la3
bor mobility earn monotonically increasing post-ranking stock returns, even after adjusting
for firm characteristics such as size, book-to-market, and past returns. Returns of zeroinvestment portfolios long high-mobility and short low-mobility stocks earn significantly
positive excess returns. The relationship between labor mobility and cross-sectional predictability of stock returns is confirmed in standard Fama and MacBeth (1973) (henceforth
FMB) cross-sectional regressions of firm-level returns on a lagged measure of labor mobility, in addition to firm-level controls. To disentangle financial leverage from labor mobility,
I repeat portfolio sorts for unlevered stock returns as a proxy for asset returns and find
similar results. As predicted by the model, the effect of labor mobility on stock returns is
stronger for highly cyclical industries, as proxied by the correlation between the industry’s
revenues and GDP.
The model assumes an exogenously given pricing kernel and is therefore silent on the true
source of aggregate risk for the economy. Although outside the scope of the model, a natural
empirical question is whether traditional asset-pricing models are able to price portfolios of
stocks sorted on the measure of labor mobility. The CAPM and Fama and French (1993)
three-factor models are rejected in Gibbons, Ross, and Shanken (1989) (GRS) tests of
alphas being jointly zero at the 1% level. The rejection of these models is economically
significant: the top labor-mobility quintile portfolio formed either at the firm- or industrylevel earn unexplained returns of 28 to 55 basis points per month, or 3.4% to 6.8% per
year. Moreover, unexplained stock returns are in general increasing in labor mobility. I
also find evidence that labor mobility is a priced risk. Using FMB regressions, I find that
a one standard deviation increase in labor mobility is associated with a 1.8% increase in
annual expected stock returns. None of the standard models–CAPM and Fama-French
three-factor models–can explain the dynamics of the labor-mobility spreads.
The rest of the paper proceeds as follows. Section 2 discusses literature related to
this work. Section 3 presents a simple model with perfect mobility or perfect immobility.
Section 4 extends the simple model to the general mobility case. Section 5 presents the
empirical tests, and Section 6 concludes.
4
2
Relation to existing literature
Related to this paper is a broad asset-pricing literature that explores the relation between
firms’ characteristics and predictability in the cross-section of asset returns.1 To this literature, this paper adds labor mobility as a theoretically motivated and observable industry
property that affects firm risk and expected stock returns.
My work also contributes to the literature that discusses the relation between laborinduced operating leverage and asset prices. Examples of this literature are Gourio (2007),
Danthine and Donaldson (2002), Cheng, Kacperczyk, and Ortiz-Molina (2009), and Parlour
and Walden (2007). Gourio (2007) proposes a model where labor intensity is positively
related to labor-induced operating leverage, a result consistent with the model proposed
here. Labor intensity and labor mobility are two complementary mechanisms that affect
a firm’s operating leverage. In a cyclical industry, the effect of labor mobility on firm risk
is increasing in labor intensity and, conversely, the effect of labor intensity on firm risk is
increasing in labor mobility.
Danthine and Donaldson (2002) discuss a mechanism where a counter-cyclical capitalto-labor share leads to labor-induced operating leverage in a general equilibrium setting.
In their model, wages are less volatile than profits, due to limited market participation of
workers, and firms provide labor-risk insurance to workers through labor contracts. Stable
wages act as an extra risk factor for shareholders, as markets are incomplete in their model.
Motivating the assumption of labor contracts that transfer labor risk to equity holders and
the assumption of limited market participation, Danthine and Donaldson (2002, pg. 42)
say:
“The twin assumptions of competitive labour markets and Cobb-Douglas production function imply, counterfactually, that factor [capital and labor] income
shares are constant over the short and medium terms.”
Although the statement is true in a single-industry economy, it does not hold under interindustry labor mobility. Despite being presented here in a partial equilibrium setting,
labor mobility causes the splitting of revenues between capital and labor to vary with time,
even under the standard assumptions of Cobb-Douglas production functions and perfect
competition. Labor mobility makes wages inelastic to asymmetric shocks in the economy, a
1
A small sample of this literature includes Basu (1983), Bhandari (1988), Fama and French (1992),
Chan and Chen (1991), Dechow, Hutton, Meulbroek, and Sloan (2000), Griffin and Lemmon (2002), Hou
and Robinson (2006), and Gourio (2007).
5
result that does not require assumptions about insurance through labor contracts, market
incompleteness, or limited participation.2
Cheng et al. (2009) empirically investigate whether unionization levels affect the cost
of capital across industries. They find that the cost of capital is higher for industries with
high unionization levels, and therefore lower flexibility on the demand side of labor. My
work focuses instead on the flexibility on the supply side of labor, i.e. the worker, and
its effect on asset prices. It remains to investigate a possible interaction between labor
mobility and frictions in the demand side of labor across industries, and its effect for the
cross-section of returns.
Parlour and Walden (2007) explore a contracting mechanism with moral hazard to
generate cross-sectional differences in risk sharing between workers and shareholders. Labor
contracts enforce effort through ex-post performance-based compensation that affects how
firm risk is split between workers and equity holders. In my model, labor contracts are
affected by the level of mobility of the labor supply in an industry, which in turn affects
the productivity of capital and expected stock returns.
Related to the theoretical approach of this paper are studies of the cross-section of
returns based on micro-level decisions through dynamic optimization. The main departure point from the literature is that, in my work, labor decisions made by workers affect
firm risk, whereas the determinant of firm risk are decisions made by equity holders on
investments in Berk, Green, and Naik (1999), Carlson, Fisher, and Giammarino (2004),
and Zhang (2005), or on hiring policy, as in Bazdresch, Belo, and Lin (2009).
3
A simple model of labor mobility
This section develops a simple partial equilibrium model that illustrates the economic
mechanism of labor mobility as a source of labor-induced operating leverage. Members of a
mobile labor force can enter and exit the industry without significant losses in productivity,
and my simple model focuses on the extreme cases in which labor supply is either immobile
or perfectly mobile. The next section extends the base model to the general case with
intermediate degrees of labor mobility. To keep the model tractable, and to focus on the
effect of micro-level decisions on firm risk, I follow Berk et al. (1999) and take the pricing
2
In a related paper, Berk and Walden (2010) discuss the role of labor mobility to explain endogenous
non-participation in financial markets. The authors argue that labor markets “complete” the markets,
through binding labor contracts between mobile and immobile agents.
6
kernel as exogenous. The dynamics of the pricing kernel Λ are given by
dΛt
= −rdt − ηdZtλ ,
Λt
(1)
where r > 0 is the instantaneous risk free rate, Z λ follows a standard Brownian motion,
and η > 0 is the market price of risk in the economy.
A large number of firms and workers make up the industry, and there is perfect competition in the goods and labor markets. Firms are identical within an industry and are
represented by an aggregate industry-firm with a standard constant-returns-to-scale CobbDouglas production technology. Perfect competition implies that wages equal the marginal
product of labor. Operating profits and wages are given by
Yt = (1 − α)At Lαt ,
(2)
Wt = αAt (Lt )α−1 ,
(3)
where Y denotes operating profits, W denotes wages inside the industry, L denotes labor
employed, 0 < α < 1 denotes labor intensity, and A denotes total factor productivity
(TFP). Physical capital is assumed to be fixed and normalized to unity. Labor is defined
as the sum of the labor productivities of all workers employed in the firm. TFP (A) growth
follows a diffusion process with constant drift µA > 0 and volatility σA > 0:
dAt
= µA dt + σA dZtA ,
At
(4)
where dZ A is the shock to TFP (A), and follows a standard Brownian motion with E dZ A dZ λ =
ρdt. I assume that ρ > 0, so that shocks to TFP growth are pro-cyclical.
Note that labor L may vary over time. The level of labor employed in the industry is
based on two state variables: TFP level (A) and wages (W ) inside the industry. While the
TFP levels are identical in the immobile and fully mobile industry, wages might not be.
In the immobile industry, labor supply is fixed and wages are given by market clearing of
workers inside the industry. Operating cash flows of the immobile industry (denoted by
“I”) are given by
YtI = (1 − α)At (LS )α ,
7
(5)
where LS is total labor employed in the immobile industry.3 From Ito’s lemma, the dynamics
of operating cash flow growth in the immobile industry are given by
dYtI
= µA dt + σA dZtA .
YtI
(6)
Note that operating cash flows are solely exposed to the industry TFP growth shock dZ A .
In the case with perfect mobility, workers can choose to work either inside or outside
the industry, continuously seeking higher wages. Labor supply is assumed unlimited, and
economy-wide wages are exogenously given and follow a geometric Brownian motion with
dynamics given by:
dWtG
= µG dt + σG dZtλ ,
WtG
(7)
where µG > 0 and σG > 0 are the drift and diffusion terms of the external-wage growth
process.4 For simplicity, economy-wide wage growth is perfectly positively correlated to
the systematic shock, so that wages are pro-cyclical.5 Labor markets are in equilibrium
when wages in the industry equal wages outside the industry, W = W G .6 Perturbations
away from this equilibrium lead to frictionless inflows or outflows of workers that equalize
wages. Labor employed in production is no longer fixed as in the immobile case, as firms
adjust their labor demand so that the marginal product of labor equals external wages.
Labor employed in the fully mobile industry (denoted by “M”) is given by:
LM
t =
At α
WtG
1
1−α
.
(8)
Plugging (8) into (2) leads to the operating profits of the industry with a perfectly mobile
3
“S” stands for “specialists”, since in the immobile industry every specialist and no generalist is employed
in the industry.
4
Here “G” stands for “generalists”, since these can be interpreted as the wages earned by generalists
workers with full mobility.
5
Relaxing this assumption does not affect the qualitative results of the paper. Moreover, the assumption
is conservative: In an extended version of the model (not presented here), I find that the effect of labor
mobility on expected returns is qualitatively decreasing in the correlation between economy-wide wage
growth and the systematic shock.
6
Here I am implicitly assuming that industry-specific shocks do not affect economy-wide wages. This is
a reasonable assumption for a sufficiently small industry definition.
8
workforce:
1
1−α
M
Yt = (1 − α)At
α
WtG
α
1−α
.
(9)
The dynamics of Y M are obtained by applying Ito’s Lemma to (9):
1
dYtM
A
λ
=
µ
dt
+
σ
dZ
−
ασ
dZ
,
M
A
G
t
t
YtM
1−α
where µM ≡
1
1−α
µA − αµG +
α
1−α
2
σA
2
+
2
σG
2
− σA σG ρ
(10)
.
The dynamics of labor employed in the industry induce differences in the exposures
to industry-specific and aggregate shocks by the owners of capital. Equation (10) shows
that on the one hand, shareholders in mobile industries are subject to a “levered” exposure
to the TFP shock Z A , where
σA
1−α
> 0 is the loading in the TFP shock, levered by the
industry’s labor intensity. On the other hand, labor costs linked to economy-wide wages
G
represent a negative exposure − ασ
< 0 to the aggregate shock Z λ . When TFP shocks are
1−α
more volatile than wage growth, the first effect dominates the second and output growth
in the full mobile industry is more volatile than that of the immobile industry, for all levels
of labor intensity. This result is formalized in the following proposition:
Proposition 1 (Labor mobility and cash flow volatility). For all levels of labor intensity
α:
σA > σG is a sufficient condition for σM > σI ,
√
where σI ≡ σA is the volatility of output growth under immobile labor and σM ≡
2 +α2 σ 2 −2ρασ σ
σA
A G
G
1−α
is the volatility of output growth under perfect labor mobility.
The Proposition follows directly from Equations (6) and (10) and the definitions of σI
and σM .
Under mild assumptions, full labor mobility also amplifies a firm’s exposure to systematic risk and leads to higher expected asset returns. To show this result, I start by deriving
the value of the firm. Unlevered asset value equals the value of the discounted stream of
operating profits:
Z
∞
V = Et
t
9
Λs
Ys ds .
Λt
(11)
The solution to equation (11) is standard, since operating cash flows follow a geometric
Brownian motion when labor is immobile or fully mobile.
Lemma 1 (Value of unlevered assets). Conditional on existing, the solution to Equation
11 for the value of a firm in the immobile and fully mobile industries is given by:
(i) The value of the fully immobile firm is given by:
VI =
YI
.
E[RI ] − µA
(12)
(ii) The value of the fully mobile firm is given by:
VM =
YM
.
E[RM ] − µM
(13)
−ασG
where E[RI ] ≡ r + ησA ρ and E[RM ] ≡ r + η ρσA1−α
, are the instantaneous expected re-
turns of the portfolio of stocks that continuously reinvests dividends, for the immobile
and fully mobile industries, respectively.
Proof: See Section A.1 of the appendix.
Unsurprisingly, the value of the firm in either industry is increasing in the TFP growth
rate, and decreasing in the interest rate r and exposure to aggregate risk, σA ρ in the
immobile case and
ρσA −ασG
1−α
in the fully mobile case.
Assumption 1 below ensures existence of a solution for the values of the firms in each
of the two cases considered (Equation 11):
Assumption 1. (Bounded values of benchmark industries)
E[RI ] − µA > 0
E[RM ] − µM > 0.
and
(14a)
(14b)
Lemma 1 shows that, under Assumption 1, the value of assets are perfectly correlated
with operating cash flows in the extreme cases of labor immobility and full mobility. Consequently, Proposition 1 can be trivially extended to show that asset return volatility is
greater in firms in the fully mobile industry. This result is formalized by the Corollary
below:
10
Corollary 1 (Labor mobility and volatility of asset returns). When Assumption 1 holds,
asset returns in the mobile industry are more volatile than those in the immobile industry:
σM > σI .
Proof: This follows directly from Proposition 1 and Lemma 1. A direct implication of
Proposition 1 is that the asset return spread between a firm in the fully mobile and fully
immobile industry is increasing in cyclicality:
Corollary 2 (Labor mobility return spread and cyclicality). Asset return spreads between
a firm in the fully mobile industry and a firm in an immobile industry are increasing in
cyclicality, ρ:
∂ (E[RM ] − E[RI ])
α
= ησA
> 0.
∂ρ
1−α
(15)
I now turn to the ranking of expected returns between a firm in the fully mobile and
fully immobile industries. Corollary 1 states that the volatility of asset returns is higher in
firms with fully mobile workforces. To extend the result for expected asset returns, I start
by making an assumption on the systematic risk loadings of the TFP and wage growth.
Let β A ≡ ρσA be the slope of a regression of TFP growth on the systematic shock Z λ , and
β G ≡ σG be the slope of a regression of economy-wide wage growth on Z λ :
Assumption 2 (Systematic risk loadings of the TFP shock and wages). TFP growth is
more exposed to systematic shocks than are economy-wide wages, β A > β G .
Assumption 2 seems plausible for most industries, since aggregate productivity is cyclical and aggregate wages tend to be smoother than TFP growth.7 Wages in a fully mobile
industry are perfectly correlated to economy-wide wages. A pro-cyclical cost per unit of labor productivity acts as a short position in systematic risk and effectively reduces expected
asset returns. When Assumption (2) holds, the decrease in systematic risk due to less elastic wages is offset by the pro-cyclical levels of labor, which act as labor-induced operating
leverage. The extra loading on systematic shocks is translated into higher expected asset
returns in mobile industries. This result is formalized below:
Proposition 2 (Ranking of asset returns in immobile and mobile industries). For all levels
of labor intensity α, β A > β G is a sufficient and necessary condition for E[RM ] > E[RI ].
7
See for example Abraham and Haltiwanger (1995) and Gourio (2007).
11
The Proposition follows from Lemma 1, and from the definitions of E[RI ] and E[RM ].
A direct implication of Lemma 1 and Proposition 2 is that the labor mobility asset
return spread between a firm in the fully mobile and a firm in an immobile industry is
increasing in the labor intensity of the productive technology. This result is formalized in
the corollary below:
Corollary 3 (Labor mobility return spread and labor intensity). Assumption 2 implies
that the asset return spread between a firm in the fully mobile industry and a firm in an
immobile industry is increasing in labor mobility, α:
∂ (E[RM ] − E[RI ])
η (σA ρ − σG )
=
> 0.
∂α
(1 − α)2
4
(16)
General model of labor mobility
This section extends the simple model with perfect labor mobility/immobility to the case
with intermediate levels of mobility. I show that, under the same assumptions used to
derive Proposition 2, expected asset returns are monotonically increasing in a measure of
labor mobility. Before establishing this result, I start by introducing labor mobility in the
general setting.
In practice, the flexibility of a worker to seek employment across industries can be
affected by a number of endowed characteristics or past decisions, such as education, age,
geographical location, and bundle of acquired labor skills. In this paper, I focus on the latter
and consider two types of workers with different sets of skills: specialists and generalists.
Specialists possess all labor skills demanded by the industry, but lack some skills demanded
elsewhere. Generalists have the full set of skills demanded outside the industry, but not
some of the skills demanded in the industry. Differences in the skill set of specialists
and generalists lead to differences in their relative labor productivity inside and outside the
industry. I define below the measure of labor mobility used in the general model, motivated
by differences in the skill set of specialists and generalists. For simplicity, I assume that
the labor productivity of both specialists working in the industry and generalists working
outside the industry is normalized to one:8
Definition 1 (Measure of labor mobility (δ)). Specialists that choose to exit and generalists
8
This can be interpreted as specialists and generalists having the same level of endowed labor skills.
12
that choose to enter the industry effectively reduce their labor productivity to
1
0 < e− δ < 1,
(17)
where δ > 0 is the measure of inter-industry labor mobility.
Figure 1 illustrates how differences in skills between the two types of workers make specialists relatively more productive inside the industry and generalists relatively more productive outside the industry. The simple model discussed in the previous section presents
the special cases of δ → 0 (immobile industry) and δ → ∞ (fully mobile industry). When
δ → 0, generalists cannot be employed in the industry and specialists cannot work elsewhere, so labor supply is immobile. When δ → ∞, specialists and generalists are effectively
identical, so that labor supply is fully mobile.
Note that this modeling choice implies symmetry in the productivity losses for specialists and generalists that exit and enter the industry, respectively. Furthermore, losses in
productivity for workers that move are fully reversible. Labor productivity is fully restored
when a worker returns to the original location. Flows of workers across industry have only
transitory effects for the firm as well. The ability of workers to move, on the other hand,
has a persistent effect on the riskiness of the firm.
Flows of workers across an industry’s border are driven by differences in wages inside
and outside the industry, and affected by labor mobility. A specialist is willing to exit and
a generalist is willing to enter the industry when the absolute difference between inside and
outside wages is high enough to justify the losses in productivity. The discrete nature of the
differences in productivity of workers inside and outside the industry creates three different
regimes of labor supply mobility: net outflow of specialists (or simply outflow ), net inflow
of generalists (or simply inflow ), and stasis regime. The outflow regime is characterized
by states where economy-wide wages sufficiently higher than wages inside the industry,
so that some specialists are willing to exit the industry. Conversely, the inflow regime is
characterized by states where wages inside the industry are relatively high, so that some
generalists are willing to enter the industry. In states where wage differentials are not high
enough to justify the decrease in productivity, there is no mobility and labor supply is
constant and entirely composed by specialists (stasis regime).
It is convenient to define a state variable as a linear transformation of the natural log
of the ratio of wages between specialists and generalists under no mobility. Let W S be the
wage earned by a specialist in the case with immobility of labor:
13
Definition 2 (Relative performance of the industry).
α
x≡
ln
1−α
The ratio
WS
WG
WS
WG
(18)
can be interpreted as the relative productivity (or performance) of the
industry relative to the –not explicitly modeled– economy-wide productivity or output. It
is easy to show that x can also be represented as the logarithm
M of the ratio of cash flows
of the fully mobile and fully immobile industries, x ≡ ln YY I . Ito’s Lemma gives the
dynamics of xt :
α
σA dZtA − σG dZtλ ,
1−α
σ2
σ2
(µG − 2G ) − (µA − 2A ) .
dxt = µx dt +
where µx ≡
α
(α−1)2
(19)
The state variable x determines the location of the industry within the labor mobility
regimes as follows. The boundary between the outflow and stasis regimes is determined
by the state where the marginal specialist is indifferent between staying and leaving the
industry. In this case, wages inside the industry are equalized to outside wages net of mo
1
α
bility frictions for specialists, WtS = e− δ WtG , or equivalently when x = xL (δ) ≡ − 1δ 1−α
.
The boundary between the stasis and inflow regimes is determined by the state where the
marginal generalist is indifferent between staying and leaving the industry. In this case,
when wages inside the industry are equalized to outside wages net of mobility frictions for
1
α
generalists, WtS = WtG e δ , or equivalently when x = xH (δ) ≡ 1δ 1−α
. To keep the notation
simple, I hereafter refer to the boundaries simply as xL and xH , although it is important to
keep in mind that they depend on the labor mobility measure δ. The table below summarizes the mobility regions in terms of the relative performance of the industry x, and the
boundaries xL and xH :
Mobility
Relative wages between
regime
specialists and generalists
Net outflow of specialists
x ≤ xL
Stasis
xL < x < x H
Net inflow of generalists
x ≥ xH
14
4.1
Cash flows
The dynamics of cash flows in each of the three regimes are related to the dynamics of cash
flows of firms in the otherwise identical fully mobile and immobile industries. In the outflow
and inflow labor regimes, a firm’s cash flows are perfectly correlated to those of a firm in an
otherwise identical fully mobile industry, since there are no associated irreversible frictions.
In the stasis regime, cash flows are identical to those of a firm in an immobile industry.
This correspondence between the regimes and the extreme mobility cases suggests that we
can use the relative performance of the fully mobile and immobile industries as a state
variable that defines the locus of the industry within the regimes.
Note that, by construction, operating cash flows of the fully mobile industry can be
solely represented as Y M = Y I ex . From the definitions of the lower and upper boundaries,
xL and xH , we can determine cash flow levels:



Y I ex+xH , if x ≤ xL


Y = Y I , if xL < x < xH



Y I ex+xL , if x ≥ x
H
(outflow of labor),
(stasis),
(20)
(inflow of labor).
Panel A in Figure 2 illustrates the relation between cash flows and the industry’s performance relative to the economy. When the industry’s performance is relatively low (small
x), cash flows are greater than those of the fully mobile industry, but smaller than those
of an otherwise identical immobile industry. When relative performance is about average,
labor mobility does not affect industry’s cash flows. When performance is relatively high
(large x), cash flows in the industry are greater than those of the immobile industry, although smaller than that of the fully mobile industry. When there is a net inflow of labor,
labor mobility increases labor supply and boosts production, but not as much as in the
case with perfect mobility.
From Equation (20), cash flow volatility in each region is given by:



σ , if x ≤ xL (outflow of labor),

 M
σ = σI , if xL < x < xH (stasis),



σ , if x ≥ x (inflow of labor),
M
H
(21)
where σM and σI are the volatility of cash flow growth of the fully mobile and immobile
industries respectively, previously defined in Proposition 1. When the volatility of the
15
TFP growth is greater than the volatility of economy-wide wage growth (σA > σG ), cash
flow growth in a fully mobile industry is more volatile than that of an immobile industry
(σM > σI ), as presented in Proposition 1. Proposition 3 below extends Proposition 1 for
the general mobility case:
Proposition 3 (The volatility of cash flow growth is weakly increasing in labor mobility).
For all levels of labor intensity α, σA > σG is a sufficient condition for
dσY
dδ
≥ 0.
Proof: The Proposition follows directly from Equation (21) and Proposition 1.
Panel B in Figure 2 illustrates Proposition 3. Cash flow growth volatility equals that
of a firm in an immobile industry in the stasis region, and that of a fully mobile industry
in the outflow and inflow of labor regions. As labor mobility increases, the stasis region
shrinks, leading to a weakly increasing volatility of cash flow growth for all values of x.
4.2
Asset returns
In this section, I show that the positive relation between labor mobility and expected asset
returns also holds in the general model. The main intuition for the result is analogous to
the case presented in the simple model: the increase in cash flow volatility due to labor
mobility adds systematic risk to the firm when economy-wide wages are less exposed to
systematic risk than is TFP (A) growth.
I start again by deriving the value of the firm, given by Equation 11. Note that the
main departing point from the simple model is that now cash flows do not follow geometric
Brownian motion. Under mild regularity conditions, we can represent Equation (11) as the
solution to the ordinary differential equation presented by Lemma 2 below:
Lemma 2 (Homogeneity of value of the unlevered firm). If the function f (x) exists and is
twice continuously differentiable, then:
(i) Equation (11) can be expressed as V = Y I f (x).
(ii) f (x) solves the following ordinary differential equation (ODE):
0 = 1 + max ex+xL − 1, 0 − max 1 − ex+xH , 0 + c1 f (x) + c2 f 0 (x) + c3 f 00 (x), (22)
where the constants c1 − c3 are given in the appendix.
16
Proof: See Section A.2 of the appendix.
Lemma 2 greatly simplifies the problem by reducing a Partial Differential Equation
(PDE) into an ODE with known closed form solution. Proposition 4 presents the solution
of the value of the firm:
Proposition 4 (Value of the unlevered firm). The solution to equation (11) is given by:



V M exH + V I B1 (pb,L (x) − pb,H (x)) , if


V = V I (1 + B3 pa,L (x) − B1 pb,H (x)) , if



V M exL + V I B (p (x) − p (x)) , if
3
a,L
a,H
x ≤ xL
(outflow of labor),
xL < x < x H
x ≥ xH
(stasis),
(23)
(inflow of labor),
where V M and V I are the values of otherwise identical firms in the fully mobile and immobile
industries (presented in Equations (13) and (12)), pb,L (x) and pb,H (x) are the prices of
claims that pay $1 when x reaches xL and xH from below, and pa,L (x) and pa,H (x) are the
prices of claims that pay $1 when the barrier is hit from above. B1 and B3 are positive
constants given in the appendix.
Proof: See Section A.3 of the appendix.
Proposition 4 shows that the value of the firm can be replicated with securities that
mimic the asset value of otherwise identical firms in fully mobile and immobile industries,
in addition to option-like securities related to each of the four barrier derivatives.
Panel A in Figure 3 illustrates Proposition 4. The figure shows the relation between
relative performance of the industry x (keeping the TFP (A) level fixed) and the value of
firms in industries with different levels of labor supply mobility (δ). In the immobile case,
the value of the firm is not affected by x for a fixed value of A. In the fully mobile case, the
value of the firm is linearly affected by ex , and the slope is determined by the labor intensity
in the industry. Even for very low strictly positive levels of δ, the value of the firm becomes
highly sensitive to x when the industry underperforms relative to the economy. When the
industry overperforms, the sensitivity of the value of the firm to x increases with δ, but not
as much as in the fully mobile case. Panel B in Figure 3 shows the value of the firm for
different values of labor mobility and relative performance of the industry. Labor mobility
decreases the value of the firm, except when the industry is over performing relative to the
economy. The figure seems to confirm the intuition that labor mobility is increases risk by
making labor supply time-varying, although it allows higher level of production when the
industry is out performing the economy, and therefore attracting workers from outside.
17
Expected unlevered asset returns are defined as the instantaneous drift of the gain
process growth:
dV + Y dt
.
E[R] ≡ E
V
(24)
As is the case for the value of the firm, the functional form for the expected returns
depends on the current labor mobility region of the industry, as shown in the Lemma below:
Lemma 3 (Instantaneous expected returns). Instantaneous expected returns are given by:
E[R] = r + η (β I + βtM ) ,
where β I ≡ σA ρ, βtM ≡
α
1−α
(25)
(σA ρ − σG ) ξ(xt ; δ), and ξ is a positive function of x given in the
appendix.
Proof: See Section A.4 of the appendix.
Equation (25) shows that the exposure to systematic risk, can be decomposed into two
components, β I and βtM . β I is the exposure to systematic risk of an immobile firm, due to
the properties of its total factor productivity. βtM is the additional exposure to systematic
risk due to labor mobility.
βtM provides the main intuition of the effect of labor mobility on asset prices. It is helpful
to consider its two main components: ξ and
α
1−α
(σA ρ − σG ). ξ captures the level of labor
mobility and is affected by the industry’s relative performance x and by δ.
α
σ ρ
1−α A
is the
exposure to systematic risk that arises from the pro-cyclical changes in labor employed,
α
and countercyclical wages induced by pro-cyclical TFP shocks. The term − 1−α
σG is the
opposing effect of economy-wide wage shocks on labor levels and wages that effectively
reduces a firm’s exposure to systematic risk. When
α
1−α
(σA ρ − σG ) > 0, flows of workers
towards the industry are pro-cyclical. Proposition 5 below shows that in this case,
∂ξ(x;δ)
∂δ
>
0, so that labor mobility is positively related to expected returns in the general model:
Proposition 5 (Instantaneous expected returns and labor mobility). Instantaneous expected returns are increasing in the measure of labor mobility (δ) if the values of the benchmark industries are well-defined (Assumption 1) and the volatility of external wages is
relatively low (Assumption 2).
∂E[R]
> 0.
∂δ
18
(26)
Proof: See Section A.5 of the appendix.
Panel A in Figure 4 illustrates the relation between relative performance of the industry
x and instantaneous expected asset returns. When Assumptions 1 and 2 hold, expected
returns of firms in fully mobile industries are higher than those in immobile industries.
Under the intermediate mobility case, expected returns are lower in the stasis region and
higher in the inflow and outflow of labor regions. When x is very low or very high, the
continuous flow of workers make the dynamics, and therefore expected asset returns, of the
industry converge to those of the fully mobile industry. For any level of relative performance x, expected returns are increasing in labor mobility. Panel B in Figure 4 illustrates
Proposition 5 using a different perspective. For fixed values of x, expected returns are
weakly increasing in δ.
Panel A in Figure 5 illustrates the interaction between cyclicality and labor mobility.
The figure shows that labor mobility spreads are increasing in the cyclicality (ρ) of the
industry. In particular, for sufficiently low values of ρ where Assumption 2 is violated, the
labor mobility spread becomes negative.
Panel B in Figure 5 illustrates that the effect of labor mobility on instantaneous expected
returns is increasing in labor intensity. This is an intuitive result, since labor intensity is a
measure of the sensitivity of capital productivity to labor levels employed in production.
4.3
Labor mobility and composition of occupations
A direct test of the model would require a ranking of industries based on δ, the mobility of workers to enter and exit the industry. Unfortunately, the labor productivity loss
associated with mobility, and therefore δ, are unobservable. In this section, I show that
the composition of occupations across industries captures the cross-sectional variation in
δ. This result motivates the use of an alternative measure based on observable occupation
data as a proxy for labor mobility in the empirical tests. I fully discuss this alternative
measure in Section 5.1.1. I start by making the following assumption on the distribution of
generalists and specialists in the economy (which is not explicitly modeled in this paper).
Assumption 3 (Distribution of generalists and specialists in the economy).
1. Generalists are not significantly concentrated in any particular industry.
2. Specialists are not significantly concentrated in any particular industry other than
their home industry.
19
Let the fraction of specialists that remain inside the industry be denoted by γ and let the
inter-industry concentration of an occupation be defined as the sum of the squared fractions
of workers assigned to the occupation in each industry.9 The inter-industry concentration
of specialists and generalists is γ 2 and 0, respectively. Now let the fraction of workers in
the industry that are specialists be denoted by ω. The inter-industry concentration of the
average worker in the industry is given by Γ ≡ γ 2 w. A measure of inter-industry mobility
is simply the inverse of the measure of the concentration measure by industry.10 Lemma 4
below shows that, under Assumption 3, (E[Γ])−1 , is in fact monotonically increasing in δ:
Lemma 4 (Expected occupational concentration of the mean worker and labor mobility).
Under Assumption 3 and for finite δ, a ranking of industries sorted on (E[Γ])−1 coincides
with a ranking of industries sorted on δ.
Proof: See Section A.6 of the appendix.
4.4
Summary of empirical implications
I summarize some of the empirical implications of the labor mobility model as follows.11
1. Expected asset returns are monotonically increasing in the measure δ of labor mobility
presented in Definition 1 (Proposition 5).
2. The “labor mobility spread” is increasing in the level ρ of industry cyclicality (Corollary 2 of Lemma 1).
3. A measure of labor mobility based on the composition of occupations, introduced in
Section 4.3 and fully discussed in Section 5.1.1, leads to the same ranking of industries
as the measure δ of labor mobility (Lemma 4).
The following section explores the latter to empirically investigate the first two implications of the model.
9
Notice that this measure is analogous to a Herfindahl index of inter-industry concentration of occupations.
10
Please refer to Section 5.1.1 for the economic motivation and complete description of the construction
of the empirical measure of labor mobility.
11
The following empirical implications implicitly assume that the value of the firm is well defined (Assumption 1), that the covariance of economy-wide wages and systematic shocks is lower than that of the
industry’s total factor productivity growth (Assumption 2), and that specialists outside their home industry
and generalists are sufficiently dispersed in the economy (Assumption 3).
20
5
Empirical evidence
The model suggests a positive relationship between labor mobility and expected asset returns (Proposition 5). I empirically investigate this link by using the two standard methodologies of portfolio sorts and Fama and MacBeth (1973) (henceforth FMB) cross-sectional
regressions. I start by introducing a new measure of labor supply mobility based on the
composition of occupations across industries and motivated by Lemma 4.
5.1
5.1.1
Data
Measuring labor mobility
I obtain-industry level occupation data from 1988 to 2009 from The Bureau of Labor Statistics (BLS). The dataset provides annual breakdowns of the number of workers assigned to
a large number standardized occupations across several industries. Please refer to Section
A.7 of the appendix for a description of this dataset and the sample of occupations and
industries used in this work.
The Herfindahl index, commonly used as an indicator of the amount of competition of
firms within an industry, inspires the inter-industry occupation-level concentration measure
proposed in this paper. I construct my measure of labor mobility in two-stages: first at
the occupation level and then aggregated at the industry level. I start by classifying
occupations in terms of inter-industry mobility based on the dispersion of workers assigned
to each occupation across industries. Let empi,j,t be the number of workers employed by
industry i and assigned to occupation j at time t. The measure of concentration of workers
assigned to occupation j at time t is given by:
conj,t =
X
2
γi,j,t
,
(27)
i
where γi,j,t ≡
empi,j,t
P
i empi,j,t
represents the fraction of all workers assigned to occupation j
that are employed in industry i at time t. For example, of the 69,500 workers assigned
to the “Airline pilots, copilots, and flight engineers,” 60,900 are found in the industry
“Scheduled Air Transportation,” so that in this particular case γi,j,t =
60,900
69,500
≈ 0.88. The
intuition behind this measure is that workers assigned to occupations concentrated in a
few industries have arguably less flexibility to switch industries than workers assigned
to occupations found across several industries. Following the example, the occupation
21
“Airline pilots, copilots, and flight engineers” is found in only 10 industries in 2009 and is
highly concentrated (conj=pilots,t=09 ≈ 0.77), which suggests low inter-industry mobility of
its workers. The occupation “Network and computer systems administrators” is found in
236 industries in 2009 and is highly dispersed (conj=sys admin,t=09 ≈ 0.04), which suggests
high inter-industry mobility.
In the second stage, I aggregate the occupation-level mobility measure by industry,
weighting by the number of employees assigned to each occupation. The measure of labor
mobility of industry i is given by:
!−1
mobi,t =
X
(conj,t × ωi,j,t )
,
(28)
j
where ωi,j,t ≡
empi,j,t
P
j empi,j,t
is the fraction of workers in industry i that are assigned to
occupation j at time t. For instance, there are 412,800 workers in the “Scheduled Air
Transportation” industry in 2009, so that for the “Airline pilots, copilots, and flight engineers” occupation, ωi,j,t =
60,900
≈
412,800
12
0.15, or equivalently 15% of all workers in the industry
are assigned to that occupation.
The measure in equation (28) is standardized in order to simplify the interpretation of
the results.13 The proposed measure is negatively related the degree of segmentation of
the labor market associated to a given industry, and is specific to the supply-side of labor,
as opposed to the demand-side of labor, which is the focus of the current work. Another
desirable property of the measure is that it is immune to ex-post shocks to the productivity
or the demand of the industry. For instance, a measure based on realized flows of workers
across industries might be affected by a latent source of firm risk that also affects asset
returns.
Table I presents a list of the bottom 20 and top 20 industries ranked by labor mobility
in the year 2009, out of a total of 282 four-digit NAICS industries. A casual inspection of
Table I reveals that many industries with low mobility have firms that are in general not
publicly listed in an exchange,14 while most of the high mobility industries are. Another
12
For the curious reader: The most common occupation in the industry is “Flight Attendants,” with a
share of 23% of all workers. The industry’s labor mobility is low, 1.36 standard deviations below the mean
in 2009.
13
The cross-sectional mean and standard deviation of the labor mobility measure are zero and one at
any given year, respectively.
14
For example “Elementary and secondary schools,” “Drinking places, alcoholic beverages,” and “Offices
of dentists.”
22
observational fact is that many manufacturing industries are represented in the top 20 list,
but not in the bottom 20, while the opposite is true for service industries.
5.1.2
Measuring industry cyclicality
The model predicts that the effect of labor mobility on asset returns is increasing in the
correlation between a firm’s Total Factor Productivity growth and the aggregate shock Z Λ
(Corollary 2), and in the empirical analysis, I use a proxy for this relationship in the form
of a measure correlation between revenues in an industry and GDP, using data from the
U.S. Bureau of Economic Analysis. The procedure is as follows: I first construct series of
quarterly revenue growth values for each firm listed on Compustat, from 1988 to 2009. I
then aggregate firms at the industry level by averaging revenue growth and weighting by
lagged revenues, leading to a time-series of industry-level revenue growth. This two-step
procedure assures that the revenue growth series for the industry consistently only considers
firms that are in the industry at each point in time. The industry-level proxy for cyclicality
is the correlation between revenue growth at the industry level and quarterly GDP growth,
using the full sample period.
5.1.3
Asset returns and portfolio formation
I consider two main types of asset returns: equity returns and unlevered equity returns.
While equity returns provide results easier to compare to those in the literature, unlevered
asset returns relate more directly to the model’s predictions of the labor-mobility effect on
firm risk. The well-known negative relation between firm risk and optimal leverage ratios
implies that part of a possible effect of labor mobility on equity risk might be offset by
lower leverage ratios. Unlevered stock returns are constructed according to:
u
f
f
ri,y,m
= ry,m
+ (ri,y,m − ry,m
)(1 − levi,y−1 )
(29)
f
where ri,y,m denotes the monthly stock return of firm i over month m of year y, ry,m
denotes
the one-month Treasury bill rate at month m of year y, and levi,y−1 denotes the leverage
ratio measure, defined as book value of debt over the sum of book value of debt plus the
market value of equity at the end of year y − 1.
Each measure of returns is presented as stock returns in excess of the one-month Treasury rate or stock returns adjusted for firm characteristics known to be ex-ante predictors of
stock returns in the cross-section. I follow the methodology in Daniel, Grinblatt, Titman,
23
and Wermers (1997) and construct 125 benchmark portfolios, sequentially triple-sorted on
the previous year’s size, book-to-market, and past stock performance (momentum rank).
The procedure uses NYSE-based breakpoints in the triple sorting and constructs valueweighted portfolios to avoid overweighting very small stocks. I then subtract the returns of
the benchmark portfolios from each constituent firm’s stock returns. An adjusted return
of zero for a given stock indicates that the return is fully explained by the firms’ size,
book-to-market, and past performance.
Furthermore, I present asset returns at both the firm and industry levels, since labor
mobility is an industry-specific characteristic. There are several common methods for
forming industry portfolios sorted on labor mobility. I construct equally weighted mobility
sorts portfolios by first creating industry portfolios of value-weighted stocks, and then
equally weighting these industry portfolios into mobility-sorted portfolios; I construct valueweighted mobility portfolios by value-weighting the industry portfolios by the sum of the
lagged market values of the firms in the industry.
Section A.8 of the appendix provides a detailed explanation of the financial and accounting data used, as well as additional constructed variables.
5.2
Characteristics of industries sorted on labor mobility
Panel A in Table II reports mean characteristics of labor mobility quintiles for the full
sample of industries over the period considered. The measures of size, assets, and sales
tend to decrease over the labor mobility quintiles. The measure of education seems to
be considerably larger for quintiles with high labor mobility. Measures of risk are slightly
increasing across labor mobility quintiles. The mean book-to-market ratio is smaller, while
betas are higher, for high labor mobility quintiles. Profitability, as measured by the ratio
of earnings to assets, shows considerable variation across the mobility quintiles. Mean
leverage ratios are significantly lower for high labor mobility quintiles.
Panels B and C in Table II report mean firm-level characteristics of labor mobility
quintiles for industries with high and low cyclicality, respectively. Pairwise comparisons
across mobility quintiles shows that the mean firms in the subsample of industries with
high and low cyclicality are fairly similar. Within each of the two sub-samples, CAPM
betas are almost flat, although they are slightly increasing across labor mobility quintiles.
Although carrying greater systematic risk through slightly larger market betas, the subsample of cyclical firms have higher mean book-to-market ratios.
24
Table III further investigates the relation between the measure of labor mobility and
average industry characteristics with FMB regressions. The table shows the time-series
means of slopes of cross-sectional regressions to estimate the degree of univariate and
multivariate correlation with the characteristics; also shown are the corresponding NeweyWest standard errors with four lags. The equation tested is:
mobi,t = λ0,t +
X
λk,t Ck,i,t + i,t ,
(30)
k
where Ck,i,t is the mean industry characteristic k of industry i at year t, and k = 0 denotes
the intercept.
Panel A shows results from Fama-MacBeth (FMB) univariate regressions of the measure
of labor mobility on each of the mean industry characteristics presented in Table II, while
Panel B shows results from multivariate FMB regressions. Labor mobility seems to be
negatively associated to both industry sales and labor intensity.
Table III shows a significant negative cross-section relation between labor mobility and
financial ratios. This finding is consistent with the positive relation between labor mobility
and firm risk, predicted by the model. Since labor mobility affects both stock returns and
the volatility of cash flow growth, it should also affect optimal capital structure decisions.
In particular, when the owners of a firm face a trade-off between operating leverage and
financial leverage,15 firms in mobile industries should have lower financial leverage levels
than their counterparts with immobile labor supplies.
5.3
5.3.1
Portfolio Sorts
Single sorts on labor mobility
This section investigates the main prediction of the model: expected returns are increasing
in labor mobility in the cross-section. To do so, I first examine average monthly returns
of portfolios sorted by the lagged measure of labor mobility. All results are presented for
equally and value-weighted portfolios, constructed both at the firm and industry levels.
Panel A of Table IV presents average post-ranking excess monthly stock returns and
monthly stock returns adjusted for size, book-to-market, and past returns, of quintiles of
mobility-sorted stocks. The first four columns indicate an increasing pattern in excess
15
See for example Gahlon and Gentry (1982) for a discussion of the trade-off between these two forms
of leverage.
25
equity returns, both at the firm and industry levels, and for both equally weighted and
value-weighted returns. The H-L portfolio is rebalanced yearly, with a long position in
stocks in the highest mobility quintile and a short position in stocks in the lowest mobility
quintile. The H-L portfolio earns economically significant excess returns of between 0.27%
and 0.51% per month. t-tests using Newey-West standard errors with four lags confirm
that the spreads are significantly different from zero. The last four columns of Panel A
present the same pattern for adjusted equity returns. Spreads are significantly different
from zero, suggesting that the relation of labor mobility and stock returns is not explained
by size, book-to-market, nor past returns.
Panel B of Table IV presents evidence for the monotonicity of the returns patterns of the
mobility-sorted portfolios presented in panel A. The first two tests, the Bonferroni bounds
test and the Wolak test, are based on the null hypothesis that the pattern is monotonic.
The Bonferroni test is based on the probability of observing a t-statistic as small as the
minimal t-statistic under the null hypothesis, given the total number of tests. The Wolak
test compares the difference between unconstrained estimates of the pairwise differences in
returns between portfolios and estimates constrained to be monotonic, using bootstrapped
errors. Neither test rejects the hypothesis that the pattern of returns across quintiles is
monotonic. The last test is the Patton and Timmermann (2010) monotonicity test, based
on the null hypothesis of a non-monotonic pattern of expected returns. The test is a onesided test of the maximum difference across the quintile returns. The test rejects the null
hypothesis of non-monotonicity in all cases, except for equally weighted portfolios formed
at the firm level.16
Table IV presents a similar analysis for unlevered returns, which are more directly
comparable to the prediction of the model. When returns are adjusted for financial leverage
ratios, the level of unlevered returns decrease across labor mobility quintiles as shown in
panel A. Spreads of the H-L portfolios are positive and statistically different from zero in
all portfolio constructions. Panel B shows that neither the Bonferroni bounds tests nor
the Wolak tests are able to reject the hypothesis that the return pattern across quintiles is
monotonic. The Patton-Timmermann test provides further confirmation of monotonicity,
by rejecting the null hypothesis that the patterns are not monotonic across all types of
portfolios.
16
I would like to thank Andrew Patton and Allan Timmermann for making available the Matlab code
adapted in the monotonicity tests described above.
26
5.3.2
Double-sorts on labor mobility and cylicality
This section investigates whether the positive relation between labor mobility and expected
returns is more significant for highly cyclical industries, as predicted by the model. I
form 10 portfolios, first by grouping firms/industries into quintiles, and then by grouping
firms/industries of each quintile into “most cyclical” (MC) and “least cyclical” (LC) halves.
The model predicts that the effect of labor mobility on firm risk and asset returns should
be increasing in the degree of cyclicality. Panels A and B of Table V provide empirical
support for this prediction. The spread of the H-L portfolios for the most cyclical half is
positive and significant for all different portfolio constructions, but in general not for the
least cyclical half. The spread between the portfolios H-L for the MC and the LC group is
positive, and in some cases statistically significant. Panels C and D of Table V present the
results of the procedure above for unlevered returns. The labor-mobility spreads of the MC
groups are again larger than those of the LC groups. The spread between the portfolios
H-L of the MC and LC groups are also positive, although not statistically different from
zero in all cases, except for equally weighted portfolios formed at the firm level.
Tables V and V show that the effect of labor mobility on asset returns is also positive
and sometimes significant for the sub-sample of least cyclical firms as well. This result is
consistent with the fact that the sample of firms considered in this work, as well as in most
finance literature, is more cyclical than the average employer in the USA. For instance, the
sample does not consider small businesses and the public sector that taken together employ
over 65% of all the workforce in the USA. Employment in the public sector and in small
businesses have been less cyclical than that of the aggregate economy in the time period
considered.17
5.4
Fama-MacBeth cross-sectional regressions
I run monthly cross-sectional regressions of post-ranking firm- and industry-level returns
on the measure of labor mobility, along with average firm characteristics known to explain
expected returns: size, book-to-market ratios, one-year-lagged stock returns, leverage ratios, and CAPM market beta. I then take averages over time of each of the slopes and
correct errors using the Newey-West methodology with four lags. Results are reported in
Table VI. Across all specifications, the average slope of returns on the measure of labor
17
Sources: Bureau of Labor Statistics report “Employment in the public sector: two recessions impact
on jobs”, BLS, 2004 and Small Business Administration report “Employment dynamics: small and large
firms over the business cycle”, Monthly Labor Review, 2007.
27
mobility is positive and significantly different from zero, and is relatively unaffected by
firms’ characteristics. These results confirm the results of the portfolio sorts, presented
in Section 5.3.1. The slopes range from 10.2 to 16.1 basis points per month, and can be
interpreted as the risk premium associated with a one-standard-deviation increase in labor
mobility. In annual terms, the slopes range from 1.3% to 1.9% per year per unit of standard
deviation.
5.5
Labor mobility and standard asset pricing models
This section investigates whether the labor mobility return spread can be explained by
other risk factors. I use the slopes of the first pass of the Fama-MacBeth regressions
presented in Table VI to represent a time-series of monthly risk premia associated with
labor mobility. I run time-series regressions of monthly labor mobility premia on different
risk factors’ risk premia. The risk factors used are the market excess returns, the size and
book-to-market factor-mimicking returns described in Fama and French (1993), and the
momentum factor-mimicking returns obtained from Kenneth French’s website. Panel A of
Table VII shows that the intercept of the time-series regressions are approximately 1.6%
per year, statistically different from zero, and quantitatively similar to the slopes on labor
mobility shown in Table VI.
Panel B of Table VII shows results of similar time-series regressions including macro
variables related to the economic environment. The variables included are the current
quarter’s GDP growth rate and monthly rate of inflation, obtained from the St. Louis
Federal Reserve Economic Database (FRED), and monthly seasonally adjusted wage and
salary disbursements, obtained from the Bureau of Economic Analysis (Tables 2.7A and
B).18 The labor mobility premium does not seem to be significantly affected by the business
cycle, as proxied by the growth in either GDP or wages, although it is positively related
to inflation. This result suggests that the labor-mobility premium increases when inflation
increases. It remains to investigate whether this result is robust to other sample periods,
and if so, the mechanism by which inflation relates to labor markets and labor mobility.
Next, I investigate whether the spreads in average returns across labor-mobility quintiles
can be explained by traditional asset-pricing models. I conduct standard two-stage timeseries asset pricing tests using the CAPM and Fama and French (1993) three-factor model.
In a first stage, I run time-series regressions of excess portfolio returns on the excess market
18
The level of the intercept cannot be interpreted as the unexplained labor mobility premium (as in
Panel A of Table VII), since the macro variables do not have zero means.
28
returns (CAPM), the combination of excess market returns and SMB and HML factors
(Fama-French model). I then use the covariance matrix of residuals in a test of whether
the intercepts of the time-series regressions are jointly different from zero, as in Gibbons
et al. (1989).
Panels A and B of Table VIII report results of asset pricing tests using portfolios
of stocks of firms sorted on the measure of labor mobility. A simple inspection of the
table reveals that neither the CAPM nor Fama-French three-factor model is able to price
portfolios of high mobility stocks, regardless of whether the portfolios are equally weighted
or value-weighted. A possible explanation of this finding is that labor mobility is a priced
risk, orthogonal to the traditional risk factors used in the literature. In the model I develop
in this paper, I consider the pricing kernel to be exogenously given. In this sense, the model
is silent on this failure of the CAPM and Fama-French three-factor models. It remains an
open question the reason of the failure of the traditional asset-pricing models in pricing
mobility-sorted stocks.
Panels A and B of Table VIII show that both the CAPM and the Fama-French threefactor models are rejected using standard GRS tests, except for the CAPM when valueweighted portfolios are used. The rejection of these traditional asset-pricing models is
particularly remarkable given the relatively short sample period considered in the tests
and the use of only five portfolios. The relation between CAPM betas and labor mobility
is mixed. For equally weighted portfolios, CAPM betas increase, and they decrease for
value-weighted portfolios.
Panels C and D of Table VIII show results of analogous tests, using portfolios of stocks
of industries sorted on the measure of labor mobility. In this case, the models also fail to
price value-weighted portfolios of industries with high labor mobility. Both the CAPM and
the Fama-French three-factor models are rejected when pricing equally weighted portfolios,
although the GRS test is unable to reject these models when value-weighted portfolios are
used.
The model is silent on why traditional asset-pricing models seem to fail to price the
“labor-mobility premium”. A hypothesis is that mobile workers transfer non-insurable
systematic risk to their employers, and that traditional asset-pricing models fail to capture
this risk because we do not observe the human capital portion of aggregate wealth. An
extension of the model into a general equilibrium setting should shed light on the failure
of traditional asset-pricing models in pricing labor mobility sorted stocks.
29
6
Conclusion
I develop a production-based model of an industry where labor mobility, i.e. the ex-ante
flexibility of workers to move across industries, subjects the owners of fixed capital to
labor supply fluctuations. For cyclical firms, labor mobility translates into a form of laborinduced operating leverage and leads to higher exposure to systematic risk. The model
provides theoretical motivation for the role of labor mobility as an industry property that
predicts cross-sectional differences in expected asset returns.
Labor mobility is determined in the model by the level of specificity of labor skills
required by the productive technology employed in the industry. I construct a new simple
empirical measure for labor mobility motivated by the model. The measure is based on
the level of inter-industry dispersion of workers in a given occupation as a proxy for labor
mobility.
I show novel supporting empirical evidence for the main predictions of the model. I
find that the return spread between assets of industries with high mobility and those of
industries with low labor mobility is significant, even after controlling for characteristics
known by the literature to explain the cross-section of returns. This finding is confirmed by
Fama-MacBeth regressions. I also find that the effect of labor mobility is larger for highly
cyclical firms, a result consistent with the model.
This work suggests that labor mobility is an observable industry characteristic with new
and promising implications for asset pricing. A natural extension of this work is to consider
the aggregate implications of labor mobility. Aggregate labor mobility might help reconcile
the apparent discrepancy between the low volatility of consumption and the high volatility
of aggregate financial indexes. In particular, labor mobility should make human capital–
the discounted value of the future stream of the sum of wages in the economy–less risky
than the portion of aggregate wealth traded in financial markets. Under this hypothesis,
changes in the relative importance of mobile workers in the economy should help forecast
changes in the relative riskiness of non-financial and financial wealth and thus expected
stock returns.
30
References
Abraham, K. G. and J. C. Haltiwanger (1995). Real wages and the business cycle. Journal
of Economic Literature 33 (3), 1215–1264.
Basu, S. (1983). The relationship between earnings’ yield, market value and return for
NYSE common stocks: Further evidence. Journal of Financial Economics 12 (1), 129–
156.
Bazdresch, S., F. Belo, and X. Lin (2009). Labor hiring, investment and stock return
predictability in the cross section. Working Paper .
Berk, J. B., R. C. Green, and V. Naik (1999). Optimal investment, growth options, and
security returns. The Journal of Finance 54 (5), 1553–1607.
Berk, J. B. and J. Walden (2010). Limited capital market participation and human capital
risk. Working Paper .
Bhandari, L. C. (1988). Debt/equity ratio and expected common stock returns: Empirical
evidence. Journal of Finance, 507–528.
Carlson, M., A. Fisher, and R. Giammarino (2004). Corporate investment and asset price
dynamics: Implications for the Cross-Section of returns. The Journal of Finance 59 (6),
2577–2603.
Chan, K. C. and N. Chen (1991). Structural and return characteristics of small and large
firms. The Journal of Finance 46 (4), 1467–1484.
Cheng, H., M. Kacperczyk, and H. Ortiz-Molina (2009). Labor unions, operating flexibility,
and the cost of equity. forthcoming at the Journal of Financial and Quantitative Analysis.
Daniel, K., M. Grinblatt, S. Titman, and R. Wermers (1997). Measuring mutual fund
performance with Characteristic-Based benchmarks. The Journal of Finance 52 (3),
1035–1058.
Danthine, J. and J. B. Donaldson (2002). Labor relations and asset returns. Review of
Economic Studies 69, 41–64.
Fama, E. F. and K. D. French (2008). Dissecting anomalies. The Journal of Finance 63 (4),
1653–1678.
31
Fama, E. F. and K. R. French (1992). The Cross-Section of expected stock returns. The
Journal of Finance 47 (2), 427–465.
Fama, E. F. and K. R. French (1993, February). Common risk factors in the returns on
stocks and bonds. Journal of Financial Economics 33 (1), 3–56.
Fama, E. F. and J. D. MacBeth (1973, June). Risk, return, and equilibrium: Empirical
tests. The Journal of Political Economy 81 (3), 607–636.
Gahlon, J. M. and J. A. Gentry (1982). On the relationship between systematic risk and
the degrees of operating and financial leverage. Financial Management 11 (2), 15–23.
Gibbons, M. R., S. A. Ross, and J. Shanken (1989, September). A test of the efficiency of
a given portfolio. Econometrica 57 (5), 1121–1152.
Gourio, F. (2007). Labor leverage, firms heterogeneous sensitivities to the business cycle,
and the Cross-Section of expected returns. Working Paper .
Griffin, J. M. and M. L. Lemmon (2002). Book-to-Market equity, distress risk, and stock
returns. The Journal of Finance 57 (5), 2317–2336.
Hou, K. and D. T. Robinson (2006). Industry concentration and average stock returns.
The Journal of Finance 61 (4), 1927–1956.
Leary, M. T. and M. R. Roberts (2010). The pecking order, debt capacity, and information
asymmetry. Journal of Financial Economics 95 (3), 332–355.
Parlour, C. and J. Walden (2007, November). Capital, contracts and the cross section of
stock returns. WP, UC Berkeley.
Patton, A. J. and A. Timmermann (2010). Monotonicity in asset returns: New tests with
applications to the term structure, the CAPM and portfolio sorts. Journal of Financial
Economics 98 (3), 605–625.
Shreve, S. (2004). Stochastic calculus for Finance II: continuous-time models. New York:
Springer.
Zhang, L. (2005). The value premium. The Journal of Finance 60 (1), 67–103.
32
A
A.1
Appendix
Proof of Lemma 1
Proof. Assume that there exists a traded asset that pays a continuous stream of dividends
identical to the operating profits of the industry. The discounted value of a portfolio of a
traded asset that continually reinvests its dividends in the asset is a martingale:
0 = ΛY +
E [d(ΛV (Y, t))]
dt
(A.1)
Applying Ito’s Lemma to equation (A.1) and simplifying leads to equation (A.2):
0 = Y + rV (Y ) + (µ − ησρ)Y V 0 (Y ) +
σ 2 2 00
Y V (Y ),
2
(A.2)
where µ and σ are the drift and volatility of the firm’s operating profits, respectively. (A.2)
Y
has a known solution given by V = r+ησρ−µ
. For the industry with immobile labor, the
particular solution becomes:
VI =
YI
r + ησI ρ − µI
(A.3)
where σI ≡ σA and µI ≡ µA . For the industry with perfectly mobile labor, the particular
solution becomes:
YM
r + η(σM1 ρ + σM2 ) − µM
2
2
σA
σG
1
α
where µM ≡ 1−α µA − αµG + 1−α 2 + 2 − σA σG ρ , σM1 ≡
VM =
A.2
(A.4)
1
σ ,
1−α A
α
and σM2 ≡ − 1−α
σG .
Proof of Lemma 2
Proof. Since Y I and Y M are sufficient statistics for operating cash flows, (equation (20)),
they should also be sufficient statistics for the value of the firm, so that V = V (Y I , Y M ).
If the function V (Y I , Y M ) exists and is twice continuously differentiable, then the value of
the firm (Equation (11)) satisfies the following partial differential equation (PDE):
0 = Y I + max [Y M exL − Y I , 0] − max [Y I − Y M exH , 0]
− V r + VI Y I (µI − σI ηρ) + VM Y M (µM − (σM1 ρ + σM2 )η)
2
2
σ2
σM1 σM2
I 2 I
M 2
+ VII (Y )
+ VMM (Y )
+
+ σM1 σM2 ρ + VIM Y I Y M (σI σM1 + σI σM2 ρ), (A.5)
2
2
2
subject to:
33
(i) transversality conditions:
lim V (Y I , Y M ) = V I
Y I →0
and
lim V (Y I , Y M ) = V M ,
Y M →0
(A.6a)
(A.6b)
(ii) value-matching conditions:
V1 (Y I , Y I exL ) = V2 (Y I , Y I exL ) and
V2 (Y I , Y I exH ) = V3 (Y I , Y I exH ),
(A.7a)
(A.7b)
where V1 , V2 , and V3 are the respective values of the firm in mobility regions 1, 2,
and 3 respectively, and
(iii) smooth-pasting conditions:
∂V1 (Y I , Y M ) ∂V2 (Y I , Y M ) and
M Ix =
M Ix
∂Y M
∂Y M
L
L
Y
=Y
e
Y
=Y
e
∂V3 (Y I , Y M ) ∂V2 (Y I , Y M ) M Ix =
M Ix .
∂Y M
∂Y M
Y =Y e H
Y =Y e H
(A.8a)
(A.8b)
Assume that there exists a traded asset that pays a continuous stream of dividends
identical to the operating profits of the industry. The discounted value of a portfolio of a
traded asset that continually reinvests its dividends in the asset is a martingale (see Shreve
(2004), pg. 234):
Et [d(Λt V (YtI , YtM ))]
0 = Λt Y (Yt , Yt ) +
dt
I
M
(A.9)
Applying Ito’s Lemma to equation (A.9) and simplifying leads to equation (A.5).
Homogeneity of cash flows implies that the solution to Equation (A.5) should also be
homogeneous of degree one in Y I and Y M . Equation (A.5) can thus be simplified to an
ordinary differential equation, as stated in Lemma 2: From Equation (20), we can express
the industry operating cash flows as:
Y = Y I + max [Y M exL − Y I , 0] − max [Y I − Y M exH , 0]
= Y I 1 + max ex+xL − 1, 0 − max 1 − ex+xH , 0
(A.10)
From equation (A.10), we can see that operating cash flows Y is homogeneous of degree
one in (Y I , Y M ). This implies that the value of the firm is also homogeneous of degree one
34
in (Y I , Y M ). We can thus express the value of the firm as V = Y I f (x), so that:
Vt = 0
VI = f (x) − xf 0 (x)
VM = e−x f 0 (x)
1
VII = I (f 00 (x) − f 0 (x))
Y
e−x 00
VMM = M (f (x) − f 0 (x))
Y
1
VIM = M (f 0 (x) − f 00 (x))
Y
(A.11a)
(A.11b)
(A.11c)
(A.11d)
(A.11e)
(A.11f)
Plugging (A.11) back to (A.5) and dividing by Y I gives the ODE:
0 = 1 + max ex+xL − 1, 0 − max 1 − ex+xH , 0 + c1 f (x) + c2 f 0 (x) + c3 f 00 (x)
where:
c1 ≡ µA − r − ησA ρ
α
(σG (η − σA ρ) − σA (ηρ − σA ))
c2 ≡ µ x +
1−α
2
α
c3 ≡
σA2 + σG2 − 2σA σG ρ
1−α
A.3
Proof of Proposition 4
Lemma (2) greatly simplifies the solution to equation (A.5) and allows us to obtain a
closed-form solution for the value of the firm, as shown below:
Proof. At region 1, equation (22) becomes:
0 = ex+xH + f1 (x)c1 + f10 (x)c2 + f100 (x)c3 ,
(A.12)
with solution:
f1 (x) = e(x+xH )b2 − e(x+xL )b2
B1
− ex+xH B2 ,
c1
where
p
c22 − 4c1 c3
b1 ≡
,
2c3
p
−c2 + c22 − 4c1 c3
b2 ≡
,
2c3
−c2 −
35
(A.13)
and
c1 + (c2 + c3 ) b1
,
(c1 + c2 + c3 ) (b1 − b2 )
1
B2 ≡
.
c1 + c2 + c3
B1 ≡
and
At region 2, equation (22) becomes:
0 = 1 + f2 (x)c1 + f20 (x)c2 + f200 (x)c3 ,
(A.14)
with solution:
f2 (x) = e(x+xL )b2
B1
B3
− e(x+xH )b1
− B4 ,
c1
c1
(A.15)
where:
c1 + (c2 + c3 ) b2
,
(c1 + c2 + c3 ) (b1 − b2 )
1
B4 ≡ .
c1
B3 ≡
and
At region 3, equation (22) becomes:
0 = ex+xL + f3 (x)c1 + f30 (x)c2 + f300 (x)c3 ,
(A.16)
with solution:
f3 (x) = e(x+xL )b1 − e(x+xH )b1
B3
− ex+xL B2 .
c1
Solving the above system of equations leads to:

(x+xH )b2 −(x+xL )b2 B1

− ex+xH B2 ,
if
 e
c1
B
B
(x+x
)b
(x+x
)b
3
1
H 1
L 2
f (x) = e
−e
− B4 ,
if
c
c

 (x+xH )b1 1 (x+xL )b1 B13
e
−e
− ex+xL B2 , if
c1
(A.17)
x ≤ xL ,
xL < x < x H ,
x ≥ xH .
(A.18)
and
 x+xH
I
(x+xH )b2
(x+xL )b2 B1

Y
e
−
e
−
e
B
,
if

2
c1

 V (I, x; δ) = Y I e(x+xH )b1 Bc13 − e(x+xL (δ))b2 Bc11 − B4 ,
if



Y I e(x+xH (δ))b1 − e(x+xL (δ))b1 B3 − ex+xL B2 , if
c1
x ≤ xL ,
xL < x < xH , (A.19)
x ≥ xH .
The result follows from the definitions of V I , V M , and from the following definitions of pb,L ,
36
pb,H , pa,L , and pa,H :
pb,L (x) ≡
XL
X
−b2
,
pb,L (x) ≡
XH
X
−b2
,
pa,L (x) ≡
X
XL
b1
pa,H (x) ≡
, and
X
XH
b1
,
where X ≡ ex , XL ≡ exL , and XL ≡ exH .
A.4
Proof of Lemma 3
Proof. Instantaneous expected returns are given by:
 h
i
dV +Y M exH dt 
E
x
≤
x
,
if

L
V 
 h
i
I dt E[R] = E dV +Y
xL < x < xH , if
V

i
h


E dV +Y M exL dt x ≥ xH ,
if
V
x ≤ xL ,
(A.20)
xL < x < x H ,
x ≥ xH .
Plugging in Equations A.19 and 20 into Equation A.20, and after some algebra manipulation, we get
α
(σA ρ − σG ) ξ(x; δ) ,
(A.21)
E[R] = r + η σA ρ +
1−α
where

1−b2
Y

b +
,


 2 V E(RM )−µM
(b2 −1)b1
ξ(x; δ) = VY pa,l E(R
−
M )−µ
M



1−b1
b 1 + Y
,
V
E(RM )−µM
b1 b2
E(RI )−µI
− pb,h
(b1 −1)b2
E(RM )−µM
−
b1 b2
E(RI )−µI
1
,
(b2 −b1 )
if
x ≤ xL ,
if
xL < x < x H ,
if
x ≥ xH .
(A.22)
The equation above makes explicit the relation between conditional expected asset returns
of the general case and the unconditional asset returns of the benchmark mobility cases.
A.5
Proof of Proposition 5
Proof. A simple inspection of Equation (25) shows that
From Assumptions 1 and 2 we have that:
∂E[R]
∂δ
> 0 is equivalent to
∂ξ(x;δ)
∂δ
> 0.
E(RI ) − µA = −c1 > 0
E(RM ) − µM = −(c1 + c2 + c3 ) > 0
ρσA − σG > 0
Lemma 5. The inequalities below follow directly from the fact that σA > 0, µA ≥ 0,
37
σG > 0, µG ≥ 0, ρ ≥ 0, 0 < α < 1, and from Assumptions 1 and 2:
c3 > 0,
b1 < 0,
b2 > 1,
B1 < 0,
B3 < 0,
I
M
I
C1 ≡ E(R ) − µA + b2 ((E(R ) − µM ) − (E(R ) − µI )) > 0,
C2 ≡ E(RI ) − µA + b1 ((E(RM ) − µM ) − (E(RI ) − µI )) > 0,
(E(RM ) − µM ) > C2
(A.23a)
(A.23b)
(A.23c)
(A.23d)
(A.23e)
(A.23f)
(A.23g)
(A.23h)
I proceed to show that the latter holds in each of the three mobility regimes:
Outflow of labor regime. In this region x ≤ xL :
(>0, Lemma 5)
(>0, Lemma 5) (>0, Assumption 2)
z
}|
{
}|
{
z }| { z
xH 2b2
M
xH 1+b2 x
b2
α
(e
)
e
1
+
b
+
(e
)
(b
−
1)
Y
(1
−
b
)B
(E(R
)
−
µ
)
2
2
2
1
M
∂ξ(x; δ)
=
>0
2
∂δ
(1 − α)δ 2 (exH )1+b2 ex + (exH )2b2 − 1 Y b2 B1 (E(RM ) − µM )
(A.24)
Stasis regime. In this region xL < x < xH :
∂ξ(x; δ)
= a(b2 − b1 )×
∂δ
(>0, Lemma 5)
(>0, Lemma 5)
z
}|
{
}|
z
{
xH (b1 +2b2 )+xb1
b2 (x+xH )
M
2
b1 (x+xH )
M
2
C2 e
(E(R ) − µM ) b1 + C1 e
(E(R ) − µM ) b2 − C2 e
(b1 + b2 )
2
(C1 exb2 − exH b2 (C2 eb1 (x+xH ) + (E(RM ) − µM ) (b1 − b2 ))) δ 2
(A.25)
Inflow of labor regime. In this region x ≥ xH :
(<0, Lemma 5)
(<0, Lemma 5)
}|
{
z }| {
xH 2b1
xH 2b1
xH 1+b1 x
b1
α
(e
)
e
(e
)
−
1
+
b
+
(e
)
b
Y
(1
− b1 )B3 (E(RM ) − µM )
1
1
∂ξ(x; δ)
>0
=
2
∂δ
b1 x
2b1
2
x
x
x
b
M
1
H
H
H
(e ) − 1 Y B3 (E(R ) − µM )
(1 − α)δ (e ) e + e
z
(A.26)
38
>0
A.6
Proof of Lemma 4
Proof. The logarithm of the ratio of cash flows of the fully mobile and fully immobile
industries x is not observable in the data, so assume that the priors for the mean and
variance of x be E[x] = 0 and Var[x] = σ 2 , for any given industry. The unconditional
expectation of Γ for a given industry is:
Z
xL
σ
e
E[Γ] =
2 zσ
α
−∞
Z
φ(z)dz +
xH
σ
xL
σ
Z
φ(z)dz +
∞
xH
σ
zσ
e− α φ(z)dz,
(A.27)
where φ is the standard normal probability density function. The empirical measure of
labor mobility is based on (E[Γ])−1 . Under Assumption 3 and for finite δ, the following
strict inequality holds:
x
2x
xH
H
− αH
α − 1
α
e
φ xσH
αe
1
+
2e
−1
1
∂(E[Γ] )
=
>0
(A.28)
∂δ
(E[Γ])2
(1 − α)σδ 2
The expression above shows that (E[Γ])−1 is in fact monotonically increasing in δ.
A.7
Occupations data
The Bureau of Labor Statistics (BLS), through its Occupational Employment Program
conducts a survey that tracks employment across occupations and industries from 1988
to 1995 and from 1997 to 2009. The survey covers approximately 200,000 non-farm establishments every six months, not including self-employed workers, from all states in the
United States. Before 1996, BLS collected data through three-year survey cycles, so that
each industry was updated every three years. After 1997 the frequency increased to annual
cycles. In the earlier sample period, I use the same industry data for three years until it is
updated by the survey to ensure a full set of industries when constructing the occupation
concentration measure.
Before 1999, the survey used the Occupational Employment Statistics (OES) taxonomy
system with 770 occupations. After 1999, the survey switched to the Standard Occupational
Classification (SOC) system with 821 to 965 occupations. Industries are classified using
three-digit Standard Industrial Classification (SIC) codes until 2001 and four-digit North
American Industry Classification System (NAICS) codes after that. I reclassify OES codes
to SOC codes using the crosswalk provided by the National Crosswalk Service Center.
This reclassification allows the merger of occupation data and data on education levels by
occupation from O*NET. Education level is the average required education per occupation
across industries, weighted by the number of employees in each occupation.
39
I exclude from the sample industries denoted as “Not Elsewhere Classified” or “Miscellaneous” (SIC xx9 and NAICS xxx9), since firms in these industries might not hold enough
similarities to justify free internal mobility of workers as in the model.
A.8
Financial and accounting data
Monthly common stock data is from the Center for Research in Security Prices (CRSP
shrcd =10 or 11). Accounting information is from Standard and Poor’s Compustat annual
industrial files. I include only stocks listed on NYSE, AMEX, and NASDAQ, with available
data in the CRSP-COMPUSTAT merged files. I follow the literature and exclude firms
with primary standard industrial classifications between 4900 and 4999 (regulated) and
between 6000 and 6999 (financials). I exclude firm-year observations with missing monthly
returns in the year and/or with missing measures of size, book-to-market, and leverage
from the previous year. Firm-level accounting variables and size measures are Winsorized
at the 0.5% level in each sample year to reduce the influence of possible outliers. For the
same reason, I exclude from the sample the lowest 5th size percentile of the sample of firms
to avoid anomalies driven by micro-cap firms, as discussed by Fama and French (2008).19
Size is defined as the market value of equity (Compustat fields prc times shrout). Book
value is defined as shareholders’ equity (Compustat SEQ) divided by the market value of
equity. I require the measures of book-to-market and size to be available at least seven
months prior to the test year. Leverage ratios are calculated as the book value of debt
adjusted for cash holdings, as reported in Compustat, divided by the assets (book-valued
leverage ratio), or divided by the sum of market value of equity and book value of debt
(market-valued leverage ratio). Tangibility ratios are defined as Property, Plant, and Equipment (PPE) divided by total assets. Profitability is defined as the ratio of earnings to total
assets. Labor intensity is proxied by the ratio of the number of employees divided by PPE.
Altman’s Z-score proxies for the likelihood of bankruptcy of a firm in the following two
years and is estimated as in Leary and Roberts (2010).20
Market betas are constructed following Fama and French (1992). Monthly market
returns are defined by the value-weighted return on all NYSE, AMEX, and NASDAQ stocks
minus the one-month Treasury bill rate. I first estimate pre-ranking betas on 60 monthly
19
The results discussed in this paper are robust to keeping all micro-caps and to excluding the lowest
10th and 20th size percentiles of firms in the sample.
20
The Z-score is defined as the sum of 3.3 times earnings before interest and taxes, plus sales, plus 1.4
times retained earnings, plus 1.2 times working capital divided by total assets.
40
returns for individual stocks. I form 100 portfolios of stocks each year, double sorted on 10
lagged size groups and 10 lagged pre-ranking market beta groups. Size groups are defined
by NYSE-based breakpoints to prevent overweighting very small stocks. I then estimate
betas for each of the portfolios using the full sample period and assign the respective beta
to each stock in the portfolio in each year. Market betas are estimated as the sum of the
slope coefficients of regressions of excess returns on contemporaneous and lagged market
excess returns.
41
Labor skills required inside industry
Inside
industry
specialist
Outside
iindustry
d t
generalist
Labor skills required outside industry
Inside
industry
generalist
Outside
industry
specialist
e

1

Figure 1. Labor productivity for specialists and generalists working inside and
outside the industry.
42
Panel A: Cash Flows
3
2.5
Immobility
Intermediate mobility
Full mobility
cash flows (Y)
2
1.5
1
0.5
0
0
0.5
1
1.5
2
x
industry relative performance (e )
2.5
3
Panel B: Cash Flow Growth Volatility
0.22
Immobility
Intermediate mobility
cash flow growth volatility
0.21
0.2
Student Version of MATLAB
0.19
0.18
0.17
0.16
0.15
0
0.5
1
1.5
2
industry relative performance (ex)
2.5
3
Figure 2. Model solution: cash flows, cash flow volatility, and industry relative
performance. Parameters values used in plots: Y I = 1, η = 0.4, r = 2.5%, α = 0.65,
LI = 1, µA = 3%, σA = 15%, ρ = 0.8 µG = 2%, and σG = 3%. Intermediate mobility case
δ = 3.
43
Student Version of MATLAB
Panel A: Unlevered Asset Values by Relative Performance Level
70
value of unlevered assets
60
Immobility
Intermediate mobility
Full mobility
50
40
30
20
10
0
0
0.5
1
1.5
2
x
industry relative performance (e )
2.5
3
Panel B: Unlevered Asset Values by Labor Mobility Level
40
value of unlevered assets
35
Low performace
Intermediate performance
High performance
Student Version of MATLAB
30
25
20
15
10
0
0.5
1
labor mobility (δ)
1.5
2
Figure 3. Model solution: unlevered asset values and earnings-over-price ratios for different levels of industry relative performance and labor mobility.
Parameters values used in plots: Y I = 1, η = 0.4, r = 2.5%, α = 0.65, LI = 1, µA = 3%,
σA = 15%, ρ = 0.8 µG = 2%, and σG = 3%. Panel A: low mobility δ = 1, and high
mobility δ = 3. Panel B: Low performance ex = 0.25, intermediate performance ex = 1,
and high performance X = 4.
44
Student Version of MATLAB
Panel A: Instantaneous Expected Returns by Labor Mobility Level
instantaneous expected returns
0.14
0.13
0.12
0.11
0.1
0.09
Immobility
Low mobility
High mobility
Full mobility
0.08
−5
0
industry relative performance (x)
5
Panel B: Instantaneous Expected Returns by Relative Performance Level
0.13
instantaneous expected returns
0.12
Low performace
Intermediate performance
High performance
Student Version of MATLAB
0.11
0.1
0.09
0.08
0
0.5
1
labor mobility (δ)
1.5
2
Figure 4. Model solution: Instantaneous expected returns for different levels
of labor mobility and industry relative performance. Parameters values used in
plots: Y I = 1, η = 0.4, r = 2.5%, α = 0.65, LI = 1, µA = 3%, σA = 15%, ρ = 0.8
µG = 2%, and σG = 3%. Panel A: low mobility δ = 1, and high mobility δ = 3. Panel
B: Low performance ex = 0.25, intermediate performance ex = 1, and high performance
X = 4.
45
Student Version of MATLAB
Panel A: Labor Mobility Spread by Industry Cyclicality Level
0.15
Labor mobility spread
instantaneous expected returns
0.1
0.05
0
−0.05
−0.1
−0.15
−0.2
−1
−0.5
0
Cyclicality (ρ)
0.5
1
Panel B: Instantaneous Expected Returns by Labor Intensity Level
0.115
instantaneous expected returns
0.11
Low labor intensity
Medium labor intensity
High labor intensity
0.105
Student Version of MATLAB
0.1
0.095
0.09
0.085
0.08
0.075
0
0.5
1
labor mobility (δ)
1.5
2
Figure 5. Model solution: Cyclicality, labor intensity, and labor mobility.
Parameters values used in plots: Y I = 1, η = 0.4, r = 2.5%, α = 0.65, LI = 1, X = 1,
µA = 3%, σA = 15%, ρ = 0.8 µG = 2%, and σG = 3%. Panel B: Low labor intensity
α = 0.55, medium labor intensity α = 0.65, and high labor intensity α = 0.75.
46
Student Version of MATLAB
Table I
Top Immobile and Top Mobile Industries
The table presents the bottom 20 (Panel A) and top 20 (Panel B) four-digit NAICS industries sorted on the measure of labor mobility, as of
2009. The measure of labor mobility is constructed as the average of the measure of occupational industry-dispersion across industries and
across time, weighted by the number of employees in each occupation.
Panel A: Bottom 20 Industries by Labor Mobility
NAICS
47
812200
812100
482100
611100
722100
561500
481100
722400
115100
722200
621200
238200
238100
561600
721100
541100
812300
611600
481200
213100
212100
515100
316200
611200
622100
Name
Death care services
Personal care services
Rail Transportation
Elementary and secondary schools
Full-service restaurants
Travel arrangement and reservation services
Scheduled air transportation
Drinking places, alcoholic beverages
Support activities for crop production
Limited-service eating places
Offices of dentists
Building equipment contractors
Building foundation and exterior contractors
Investigation and security services
Traveler accommodation
Legal services
Drycleaning and laundry services
Other schools and instruction
Nonscheduled air transportation
Support activities for mining
Coal and petroleum gases
Radio and television broadcasting
Footwear
Junior colleges
General medical and surgical hospitals
Panel B: Top 20 Industries by Labor Mobility
mobility
NAICS
-1.54
-1.52
-1.52
-1.47
-1.46
-1.43
-1.39
-1.37
-1.35
-1.35
-1.32
-1.29
-1.29
-1.29
-1.29
-1.19
-1.16
-1.12
-1.11
-1.06
-1.05
-1.04
-1.03
-1.02
-1.01
332800
335200
332500
423500
325500
325600
336100
424600
312200
332600
424800
424100
332200
327100
425100
327200
336500
423800
322200
335100
493100
424300
423300
333100
311200
Name
Coating, engraving, and heat treating metals
Household appliances and miscellaneous machines
Hardware manufacturing
Metal and mineral merchant wholesalers
Paint, coating, and adhesive manufacturing
Soap, cleaning compound, and toiletry mfg.
Motor vehicles
Chemical merchant wholesalers
Tobacco products
Spring and wire product manufacturing
Alcoholic beverage merchant wholesalers
Paper and paper product merchant wholesalers
Cutlery and handtool manufacturing
Clay product and refractory manufacturing
Electronic markets and agents and brokers
Glass and glass product manufacturing
Railroad rolling stock
Machinery and supply merchant wholesalers
Converted paper product manufacturing
Electric lighting equipment manufacturing
Warehousing and storage
Apparel and piece goods merchant wholesalers
Lumber and construction supply merchant wholesale
Agriculture, construction, and mining machinery mf
Grain and oilseed milling
mobility
2.39
2.26
2.08
2.06
1.99
1.94
1.91
1.84
1.84
1.76
1.75
1.73
1.72
1.70
1.69
1.65
1.62
1.62
1.61
1.60
1.59
1.58
1.50
1.46
1.45
Table II
Summary Statistics
The table reports time-series averages of firm-level characteristics of industries sorted by labor mobility. The table
shows statistics for all industries in the sample (Panel A), for the subsample of industries with correlation between
quarterly revenues and GDP above median (Panel B), and below median (Panel C). “Mob.” is the measure of
labor mobility, and constructed as the average of the measure of occupational industry-dispersion across industries
and across time, weighted by the number of employees in each occupation. “ln(Size) is defined as logarithm of
the market value of equity plus book value of total debt. “ln(Assets)” is defined as logarithm of the total book
value of assets. “ln(Sales)” is defined as logarithm of the total firm operating profits in a year. “Edu.” is the
mean education level, measured as the average required education per occupation across industries, weighted by
the number of employees in each occupation. “Lab.Int.” is labor intensity, defined as the ratio of the number of
employees divided by PPE. “B/M” is the ratio of book value, defined as shareholders’ equity divided by market
value of equity. “Prof.” is profitability, defined as the ratio of earnings to total assets. “Lever.” is leverage ratio,
defined as the ratio of book value of debt adjusted for cash holdings, as reported in Compustat, divided by the
assets. “Beta” is CAPM, constructed following the methodology described in Fama and French (1992). The sample
covers the period 1989–2009.
Quintile
Mob.
ln(Size)
ln(Assets) ln(Sales)
Edu.
Lab.Int.
B/M
Prof.
Lever.
Beta
Panel A: Characteristics of Mobility Sorted Quintile Portfolios
Low
-1.08
12.71
6.08
5.95
-0.06
0.524
0.803
0.130
0.222
1.237
2
-0.34
12.71
6.01
6.07
0.13
0.648
0.804
0.123
0.186
1.319
3
0.31
12.69
5.90
5.96
0.14
0.580
0.785
0.097
0.173
1.316
4
0.90
12.41
5.49
5.47
0.45
0.529
0.702
0.095
0.150
1.348
High
1.52
12.31
5.53
5.70
0.44
0.536
0.761
0.114
0.159
1.328
Panel B: Characteristics of Mobility Sorted Quintile Portfolios (High Cyclicality Firms)
Low
-1.13
12.94
6.33
6.39
-0.34
0.479
0.846
0.137
0.220
1.248
2
-0.34
12.59
5.87
3
0.30
12.66
6.14
5.87
0.22
0.645
0.817
0.111
0.169
1.367
6.37
-0.08
0.759
0.938
0.127
0.221
1.302
4
0.91
12.46
5.80
5.97
0.22
0.530
0.818
0.115
0.196
1.339
High
1.53
12.30
5.60
5.80
0.48
0.604
0.813
0.115
0.178
1.338
Panel C: Characteristics of Mobility Sorted Quintile Portfolios (Low Cyclicality Firms)
Low
-1.04
12.45
5.80
5.47
0.30
0.530
0.760
0.122
0.223
1.224
2
-0.35
12.87
6.20
6.32
-0.03
0.590
0.802
0.138
0.209
1.258
3
0.31
12.76
5.77
5.67
0.34
0.515
0.667
0.070
0.137
1.322
4
0.88
12.40
5.21
5.01
0.63
0.481
0.588
0.073
0.107
1.357
High
1.51
12.36
5.49
5.65
0.36
0.472
0.702
0.117
0.143
1.307
48
Table III
Fama-MacBeth Regressions of Labor Mobility on Industry Average
Characteristics
The table reports average (across years) slopes of yearly cross-sectional regressions of the measure of labor
mobility on industry average characteristics and corresponding standard errors. Variables are defined in
Table II. Panel A shows results of 11 univariate regressions (by columns) and Panel B shows results of 10
multivariate regressions (by rows). Newey-West standard errors are estimated with four lags. Significance
levels are denoted by (* = 10% level), (** = 5% level) and (*** = 1% level). The sample covers the period
1989–2009.
ln(Size)
ln(Assets)
ln(Sales)
Edu.
Lab.Int.
B/M
Prof.
Lev.
0.02
-0.16
-0.42**
(0.06)
(0.39)
(0.15)
0.02
-0.04
-0.83***
(0.06)
(0.53)
(0.14)
0.05
-0.02
-0.72***
(0.06)
(0.53)
(0.17)
0.03
-0.53
-0.96***
(0.06)
(0.49)
(0.17)
0.02
-0.04
-0.83***
(0.06)
(0.53)
(0.14)
Panel A: Simple Regressions
-0.01
-0.02
(0.01)
(0.01)
0.04***
(0.01)
-0.04
-0.15***
(0.03)
(0.03)
Panel B: Multiple Regressions
-0.05**
-0.06
-0.17***
(0.02)
(0.04)
(0.03)
-0.05**
-0.06
-0.17***
(0.02)
(0.04)
(0.03)
-0.05
-0.16***
(0.04)
(0.03)
-0.05**
-0.06
-0.17***
(0.02)
(0.04)
(0.03)
-0.01
-0.06
0.06
-0.22
-0.65***
(0.02)
(0.04)
(0.06)
(0.55)
(0.14)
-0.05***
-0.05
-0.17***
-0.03
-0.82***
(0.02)
(0.04)
(0.03)
(0.52)
(0.11)
-0.04**
-0.06*
-0.17***
(0.02)
(0.03)
(0.03)
(0.06)
-0.06***
-0.03
-0.16***
-0.09
0.04
(0.02)
(0.04)
(0.03)
(0.06)
(0.53)
-0.05**
-0.06
-0.17***
0.02
-0.04
-0.83***
(0.02)
(0.04)
(0.03)
(0.06)
(0.53)
(0.14)
0.06***
(0.02)
0.04
-0.83***
(0.14)
-0.01
-0.44***
(0.01)
(0.15)
49
Table IV
Labor Mobility and The Cross Section of Returns
The table reports post-ranking mean realized excess monthly stock returns over one-month Treasury bill
rates, and adjusted monthly returns of portfolios of firms/industries sorted on labor mobility. Panel A
shows results for returns, and Panel B reports monotonicity test results of the return patterns presented in
Panel A. These tests are described in Patton and Timmermann (2010). Panel C shows results for unlevered
equity excess returns, estimated as excess stock returns times one minus lagged leverage ratio (measured
using book value of debt and market value of equity). Panel D reports monotonicity test results of the return
patterns presented in Panel C. “Adjusted Returns” are adjusted for size, book-to-market, and momentum,
according to the methodology in Daniel et al. (1997). “Firm” and “Industry” indicate portfolios formed
at the firm and industry levels, respectively. “Equal” and “Value” indicate portfolios formed using equally
weighted and value-weighted returns, respectively. H-L is the zero investment portfolio long the portfolio
of industries with high labor mobility (H) and short the portfolio of industries with low labor mobility (L).
Newey-West standard errors are estimated with four lags. Significance levels are denoted by (* = 10%
level), (** = 5% level) and (*** = 1% level). The sample covers the period 1989–2009.
Excess Returns
Portfolio/
Test
Firm
Equal
Adjusted Returns
Industry
Value
Equal
Value
Firm
Equal
Industry
Value
Equal
Value
Panel A: Portfolio Equity Returns
L
0.73
0.40
0.57
0.34
-0.14
-0.22
-0.33
-0.26
2
0.64
0.39
0.68
0.50
-0.25
-0.24
-0.20
-0.16
3
0.95
0.55
0.74
0.56
0.04
-0.11
-0.19
-0.09
4
1.02
0.58
0.76
0.64
0.15
-0.03
-0.15
0.01
H
1.07
0.90
0.87
0.85
0.12
0.22
-0.11
0.20
H-L
0.33**
0.49**
0.30***
0.51**
0.26*
0.44***
(0.20)
(0.21)
(0.11)
(0.23)
(0.17)
0.22**
0.45***
(0.16)
(0.10)
(0.18)
Panel B: Monotonicity Tests
Bonferroni (H0: monotonicity)
reject?
N
N
N
N
N
N
N
N
p -value
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
Wolak (H0: monotonicity)
reject?
N
N
N
N
N
N
N
N
p -value
0.913
0.990
1.000
0.980
0.852
0.986
0.990
1.000
Patton-Timmerman (H0: no monotonicity)
reject?
Y
Y
Y
Y
N
Y
Y
Y
p -value
0.099
0.026
0.004
0.002
0.170
0.021
0.002
0.001
50
Table IV
Labor Mobility and The Cross Section of Returns
(continued from previous page)
Excess Returns
Portfolio/
Test
Firm
Equal
Adjusted Returns
Industry
Value
Equal
Value
Firm
Equal
Industry
Value
Equal
Value
Panel C: Portfolio Unlevered Returns
L
0.51
0.30
0.37
0.23
-0.16
-0.19
-0.30
-0.24
2
0.45
0.31
0.51
0.40
-0.24
-0.21
-0.19
-0.14
3
0.75
0.47
0.48
0.49
0.05
-0.13
-0.16
-0.11
4
0.85
0.50
0.57
0.53
0.14
0.01
-0.12
0.03
H
0.90
0.83
0.66
0.77
0.14
0.23
-0.06
0.22
H-L
0.39**
0.53***
0.29***
0.54***
0.30**
0.42***
(0.18)
(0.19)
(0.09)
(0.20)
(0.15)
0.23***
0.46***
(0.14)
(0.09)
(0.16)
Panel D: Monotonicity Tests
Bonferroni (H0: monotonicity)
reject?
N
N
N
N
N
N
N
N
p -value
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
Wolak (H0: monotonicity)
reject?
N
N
N
N
N
N
N
N
p -value
0.952
0.990
0.975
0.990
0.885
0.978
1.000
0.950
Patton-Timmerman (H0: no monotonicity)
reject?
Y
Y
Y
Y
N
Y
Y
Y
p -value
0.065
0.008
0.030
0.003
0.126
0.024
0.000
0.009
51
Table V
Cyclicality, Labor Mobility, and the Cross-Section of Returns
The table reports post-ranking mean realized excess monthly stock returns over one-month Treasury bill rates, and adjusted monthly
returns of portfolios of firms/industries double sorted on labor mobility (rows) and cyclicality (columns). Panel A shows results for
equity returns, and Panel B reports monotonicity test results of the return patterns presented in Panel A. These tests are described in
Patton and Timmermann (2010). Panel C shows results for unlevered equity excess returns, estimated as excess stock returns times
one minus lagged leverage ratio (measured using book value of debt and market value of equity). Panel D reports monotonicity test
results of the return patterns presented in Panel C. “Adjusted equity returns” and “Adjusted Unlevered Returns” are adjusted for
size, book-to-market, and momentum, according to the methodology in Daniel et al. (1997). “Firm” and “Industry” indicate portfolios
formed at the firm and industry levels, respectively. H-L is the zero investment portfolio long the portfolio of industries with high labor
mobility (H) and short the portfolio of industries with low labor mobility (L). MC-LC is the difference between returns of portfolios of
firms/industries of least and most cyclical half of each labor mobility quintile. Newey-West standard errors are estimated with four lags.
Significance levels are denoted by (* = 10% level), (** = 5% level) and (*** = 1% level). The sample covers the period 1989–2009.
Portfolio
Firm
Industry
Equally weighted
Least
Cyclical
Most
Cyclical
Value-weighted
MC-LC
Least
Cyclical
Most
Cyclical
Equally weighted
MC-LC
Least
Cyclical
Most
Cyclical
Value-weighted
MC-LC
Least
Cyclical
Most
Cyclical
MC-LC
Panel A: Excess Stock Returns of Portfolios Double Sorted by Mobility and Cyclicality
52
L
0.91
0.54
-0.37
0.48
0.26
-0.22
0.69
0.43
-0.26
0.58
0.30
-0.28
2
0.54
0.73
0.19
0.28
0.50
0.22
0.71
0.68
-0.03
0.52
0.51
-0.01
3
0.95
0.97
0.02
0.44
0.77
0.33
0.71
0.78
0.07
0.17
0.63
0.46
4
1.16
0.94
-0.22
0.79
0.38
-0.41
0.80
0.57
-0.23
0.70
0.63
-0.07
H
0.97
1.13
0.16
0.65
0.91
0.26
0.82
1.06
0.24
0.87
0.93
0.06
H-L
0.06
0.58***
0.52*
0.17
0.65***
0.48
0.14
0.63***
0.50*
0.29
0.64**
0.35
(0.29)
(0.27)
(0.18)
(0.21)
(0.33)
(0.28)
(0.43)
(0.26)
(0.24)
(0.33)
(0.41)
(0.33)
Panel B: Adjusted Equity Returns of Portfolios Double Sorted by Mobility and Cyclicality
L
-0.07
-0.24
-0.17
-0.20
-0.24
-0.04
-0.32
-0.37
-0.05
-0.18
-0.20
-0.02
2
-0.32
-0.08
0.24
-0.21
-0.06
0.15
-0.28
-0.14
0.14
-0.12
-0.07
0.05
3
0.00
0.04
0.04
-0.23
0.05
0.28
-0.25
-0.09
0.16
-0.29
-0.12
0.17
4
0.29
0.07
-0.22
0.22
-0.16
-0.38
-0.07
-0.31
-0.24
-0.01
0.05
0.06
H
0.02
0.16
0.14
-0.11
0.31
0.42
-0.14
0.09
0.23
0.15
0.43
0.28
H-L
0.09
0.41**
0.32
0.09
0.55***
0.46*
0.18
0.46***
0.28
0.33*
0.63***
0.29
(0.27)
(0.24)
(0.33)
(0.17)
(0.29)
(0.23)
(0.21)
(0.19)
(0.18)
(0.18)
(0.23)
(0.33)
Table V
Cyclicality, Labor Mobility, and the Cross-Section of Returns
(continued from previous page)
Portfolio
Firm
Industry
Equally weighted
Least
Cyclical
Most
Cyclical
Value-weighted
MC-LC
Least
Cyclical
Most
Cyclical
Equally weighted
MC-LC
Least
Cyclical
Most
Cyclical
Value-weighted
MC-LC
Least
Cyclical
Most
Cyclical
MC-LC
Panel C: Excess Unlevered Returns of Portfolios Double Sorted by Mobility and Cyclicality
53
L
0.65
0.37
-0.28
0.34
0.19
-0.15
0.43
0.29
-0.14
0.42
0.21
-0.21
2
0.38
0.52
0.14
0.22
0.43
0.21
0.52
0.43
-0.09
0.37
0.44
0.07
3
0.82
0.72
-0.10
0.39
0.62
0.23
0.56
0.51
-0.05
0.12
0.50
0.38
4
1.02
0.73
-0.29
0.72
0.26
-0.46
0.54
0.41
-0.13
0.64
0.47
-0.17
H
0.82
0.94
0.12
0.58
0.83
0.25
0.68
0.79
0.11
0.80
0.86
0.06
H-L
0.18
0.57***
0.39*
0.24
0.65***
0.40
0.25**
0.50***
0.25
0.38*
0.64***
0.26
(0.27)
(0.26)
(0.24)
(0.27)
(0.21)
(0.22)
(0.24)
(0.38)
(0.13)
(0.17)
(0.25)
(0.36)
Panel D: Adjusted Unlevered Returns of Portfolios Double Sorted by Mobility and Cyclicality
L
-0.09
-0.22
-0.13
-0.17
-0.19
-0.02
-0.33
-0.28
0.05
-0.16
-0.18
-0.02
2
-0.31
-0.11
0.20
-0.19
-0.08
0.11
-0.24
-0.12
0.12
-0.14
-0.06
0.08
3
0.04
0.03
-0.01
-0.25
0.03
0.28
-0.21
-0.05
0.16
-0.28
-0.14
0.14
4
0.28
0.06
-0.22
0.24
-0.17
-0.41
-0.11
-0.28
-0.17
-0.01
-0.01
0.00
H
0.03
0.18
0.15
-0.07
0.26
0.33
-0.04
0.09
0.13
0.19
0.39
0.20
H-L
0.13
0.40***
0.27
0.10
0.45***
0.35
0.37***
0.08
0.35**
0.57***
0.23
(0.21)
(0.21)
(0.18)
(0.16)
(0.16)
(0.29)
0.29***
(0.12)
(0.14)
(0.21)
(0.20)
(0.20)
(0.29)
Table VI
Fama-MacBeth Cross-Sectional Regressions of Labor Mobility Measure and
Monthly Returns
The table time-series means of slopes and standard errors from monthly cross-section
regressions of stock returns (in basis points) on firm-level characteristics, described in
Table II. Panel A reports results of tests based on individual stocks. Panel B reports
results from tests based on industry-level portfolios. R2 is the average adjusted r-.
squared across monthly regressions. Newey-West standard errors are estimated with
four lags. Significance levels are denoted by (* = 10% level), (** = 5% level) and (***
= 1% level). The sample covers the period 1989–2009.
Panel A: Firm Level
Mob.
ln(Size)
B/M
Mom.
Lever.
Beta
15.7**
(6.9)
R2
0.2
16.1***
(6.1)
-7.3
(5.6)
22.3*
(11.4)
-48.1*
(28.0)
14.6***
(5.1)
-7.5
(4.7)
24.6***
(7.8)
-46.9*
(25.1)
2.1
0.3
(43.5)
-48.5
(44.6)
3.4
Beta
R2
Panel B: Industry Level
Mob.
ln(Size)
B/M
Mom.
Lever.
10.5***
(3.7)
0.2
10.2***
(3.9)
-11.2*
(6.5)
14.9
(18.2)
6.3
(37.6)
10.4***
(3.9)
-13.1**
(6.4)
15.3
(16.6)
14.4
(34.5)
54
3.8
-18.6
(53.9)
0.8
(57.0)
6.4
Table VII
Time Series Regressions of Labor Mobility Measure
This table presents results from time-series regressions of the measure of labor mobility premium (in basis
points) on various asset-pricing factors and business cycle indicators. Labor mobility premium is obtained as
the slope of monthly cross-sectional regressions of stock returns on the measure of labor mobility (presented
in Table VI). Panel (A) presents regression results with risk factors. “MKT” is the market excess return,
“SMB” and “HML” are size and book-to-market factors-mimicking returns described in Fama and French
(1993), and “UMD” is the momentum factor-mimicking return obtained from Kenneth French’s website.
.
Panel (B) includes variables that proxy for the business cycle. “GDP” is the current quarter’s GDP
growth rate, obtained from the St. Louis Federal Reserve Economic Database (FRED), “WAG” is monthly
seasonally adjusted wage and salary disbursements obtained from the Bureau of Economic Analysis (Tables
2.7A and B). “INF” is the monthly rate of inflation, obtained from FRED. Newey-West standard errors
are estimated with four lags. Significance levels are denoted by (* = 10% level), (** = 5% level) and (***
= 1% level). The sample covers the period 1989–2009.
Intercept
MKT
SMB
HML
UMD
GDP
WAG
INF
Panel A: Robustness of Labor Mobility Premia and Risk Factors
14.0***
(4.8)
124.6
(106.9)
15.2***
(4.7)
-24.3
(108.7)
381.9***
(140.4)
-372.1**
(150.2)
12.3***
(4.7)
107.6
(113.8)
363.0***
(137.7)
-256.4*
(151.3)
297.9***
(90.8)
Panel B: Labor Mobility Premia and the Business Cycle
7.3
(6.6)
1177.8
(741.4)
13.8***
(5.1)
225.0
(512.9)
13.2***
(4.8)
1060.3*
(608.3)
7.5
(6.6)
11.1*
(6.4)
95.1
(116.2)
351.9**
(137.8)
-270.8*
(151.8)
285.8***
(97.1)
55
1231.0
(767.6)
-665.1
(638.1)
1389.3*
(738.9)
210.4
(779.2)
-464.6
(607.5)
1291.7*
(702.6)
Table VIII
Time-Series Asset Pricing Tests
This table reports asset pricing tests of the CAPM and Fama and French (1993) model using five portfolios
of stocks sorted on labor mobility at the firm level. The tables reports the intercept (monthly alpha) of
time-series regressions of excess portfolio returns at the firm- and industry-level on the excess market returns
(Panels (A) and (C), respectively), and excess market returns at the firm- and industry-level, and SMB and
.
HML factors (Panels (B) and (D), respectively ). GRS stands for the F-statistic for the Gibbons et al. (1989)
test of whether the intercepts of the time-series regressions are jointly different from zero, as in Gibbons et al.
(1989). Newey-West standard errors are estimated with four lags. Significance levels are denoted by (* = 10%
level), (** = 5% level) and (*** = 1% level). The sample covers the period 1989–2009.
Panel A: CAPM Tests (sorts at firm level)
Equally weighted Portfolios
L
2
3
4
Alpha (%)
0.24
(0.24)
0.11
(0.23)
0.45*
(0.25)
0.50**
(0.24)
MKT Beta
1.04
(0.08)
1.14
(0.06)
1.06
(0.05)
1.11
(0.05)
GRS
p -val (%)
3.43
0.52
Value-weighted Portfolios
H
L
2
3
4
H
0.55**
(0.24)
-0.06
(0.14)
-0.13
(0.13)
0.09
(0.18)
0.15
(0.16)
0.47***
(0.16)
1.10
(0.05)
1.00
(0.05)
1.10
(0.04)
0.99
(0.07)
0.93
(0.06)
0.92
(0.06)
2.13
6.24
Panel B: Fama-French Three Factor Model Tests (sorts at firm level)
Equally weighted Portfolios
L
2
3
4
Alpha (%)
-0.05
(0.16)
-0.08
(0.15)
0.26
(0.17)
0.34**
(0.14)
MKT Beta
1.07
(0.05)
1.06
(0.04)
1.00
(0.04)
SMB Beta
0.54
(0.11)
0.74
(0.07)
HML Beta
0.57
(0.09)
0.25
(0.07)
GRS
p -val (%)
3.72
0.29
Value-weighted Portfolios
H
L
2
3
4
0.34**
(0.14)
-0.13
(0.14)
-0.08
(0.14)
0.22
(0.16)
0.10
(0.14)
0.41***
(0.15)
1.00
(0.03)
1.03
(0.03)
1.07
(0.05)
1.07
(0.04)
0.94
(0.05)
0.95
(0.05)
0.93
(0.04)
0.68
(0.10)
0.82
(0.06)
0.74
(0.10)
-0.13
(0.07)
0.02
(0.07)
-0.09
(0.06)
0.03
(0.06)
0.09
(0.07)
0.29
(0.08)
0.18
(0.06)
0.31
(0.07)
0.21
(0.07)
-0.12
(0.06)
-0.29
(0.08)
0.11
(0.09)
0.10
(0.09)
2.48
3.28
56
H
Table VIII
Time-Series Asset Pricing Tests
(continued from previous page)
Panel C: CAPM Tests (sorts at industry level)
Equally weighted Portfolios
L
2
3
4
Alpha (%)
0.24
(0.24)
0.26
(0.23)
0.31
(0.24)
0.53**
(0.25)
MKT Beta
1.01
(0.09)
1.16
(0.05)
1.04
(0.06)
1.09
(0.05)
GRS
p -val (%)
2.21
5.40
Value-weighted Portfolios
H
L
2
3
4
H
0.53**
(0.25)
-0.10
(0.13)
-0.02
(0.13)
0.13
(0.14)
0.23
(0.14)
0.43**
(0.20)
1.04
(0.07)
0.96
(0.05)
1.13
(0.03)
0.89
(0.04)
0.90
(0.05)
0.87
(0.07)
1.38
23.43
Panel D: Fama-French Three Factor Model Tests (sorts at industry level)
Equally weighted Portfolios
L
2
3
Alpha (%)
-0.05
(0.16)
0.05
(0.13)
0.08
(0.18)
MKT Beta
1.04
(0.05)
1.10
(0.03)
SMB Beta
0.52
(0.10)
HML Beta
0.55
(0.08)
GRS
p -val (%)
2.57
2.73
4
Value-weighted Portfolios
H
L
2
3
4
0.34**
(0.14)
0.28*
(0.15)
-0.17
(0.13)
0.01
(0.13)
0.15
(0.13)
0.19
(0.13)
0.37**
(0.18)
1.04
(0.04)
0.99
(0.04)
1.02
(0.04)
1.03
(0.04)
1.10
(0.03)
0.95
(0.04)
0.91
(0.04)
0.89
(0.05)
0.75
(0.09)
0.52
(0.14)
0.87
(0.05)
0.64
(0.14)
-0.12
(0.08)
0.09
(0.05)
-0.29
(0.06)
0.06
(0.05)
0.03
(0.07)
0.32
(0.07)
0.41
(0.09)
0.24
(0.06)
0.44
(0.08)
0.20
(0.07)
-0.09
(0.05)
0.04
(0.08)
0.09
(0.08)
0.15
(0.10)
1.27
27.76
57
H